You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

2549 lines
84 KiB

8 years ago
MDEV-16678 Prefer MDL to dict_sys.latch for innodb background tasks This is joint work with Thirunarayanan Balathandayuthapani. The MDL interface between InnoDB and the rest of the server (in storage/innobase/dict/dict0dict.cc and in include/) is my work, while most everything else is Thiru's. The collection of InnoDB persistent statistics and the defragmentation were not refactored to use MDL. They will keep relying on lower-level interlocking with fil_check_pending_operations(). The purge of transaction history and the background operations on fulltext indexes will use MDL. We will revert commit 2c4844c9e76427525e8c39a2d72686085efe89c3 (MDEV-17813) because thanks to MDL, purge cannot conflict with DDL operations anymore. For a similar reason, we will remove the MDEV-16222 test case from gcol.innodb_virtual_debug_purge. Purge is essentially replacing all use of the global dict_sys.latch with MDL. Purge will skip the undo log records for tables whose names start with #sql-ib or #sql2. Theoretically, such tables might be renamed back to visible table names if TRUNCATE fails to create a new table, or the final rename in ALTER TABLE...ALGORITHM=COPY fails. In that case, purge could permanently leave some garbage in the table. Such garbage will be tolerated; the table would not be considered corrupted. To avoid repeated MDL releases and acquisitions, trx_purge_attach_undo_recs() will sort undo log records by table_id, and purge_node_t will keep the MDL and table handle open for multiple successive undo log records. get_purge_table(): A new accessor, used during the purge of history for indexed virtual columns. This interface should ideally not exist at all. thd_mdl_context(): Accessor of THD::mdl_context. Wrapped in a new thd_mdl_service. dict_get_db_name_len(): Define inline. dict_acquire_mdl_shared(): Acquire explicit shared MDL on a table name if needed. dict_table_open_on_id(): Return MDL_ticket, if requested. dict_table_close(): Release MDL ticket, if requested. dict_fts_index_syncing(), dict_index_t::index_fts_syncing: Remove. row_drop_table_for_mysql() no longer needs to check these, because MDL guarantees that a fulltext index sync will not be in progress while MDL_EXCLUSIVE is protecting a DDL operation. dict_table_t::parse_name(): Parse the table name for acquiring MDL. purge_node_t::undo_recs: Change the type to std::list<trx_purge_rec_t*> (different container, and storing also roll_ptr). purge_node_t: Add mdl_ticket, last_table_id, purge_thd, mdl_hold_recs for acquiring MDL and for keeping the table open across multiple undo log records. purge_vcol_info_t, row_purge_store_vsec_cur(), row_purge_restore_vsec_cur(): Remove. We will acquire the MDL earlier. purge_sys_t::heap: Added, for reading undo log records. fts_sync_during_ddl(): Invoked during ALGORITHM=INPLACE operations to ensure that fts_sync_table() will not conflict with MDL_EXCLUSIVE. Uses fts_t::sync_message for bookkeeping.
6 years ago
11 years ago
11 years ago
MDEV-10139 Support for InnoDB SEQUENCE objects We introduce a NO_ROLLBACK flag for InnoDB tables. This flag only works for tables that have a single index. Apart from undo logging, this flag will also prevent locking and the assignment of DB_ROW_ID or DB_TRX_ID, and imply READ UNCOMMITTED isolation. It is assumed that the SQL layer is guaranteeing mutual exclusion. After the initial insert of the single record during CREATE SEQUENCE, InnoDB will be updating the single record in-place. This is crash-safe thanks to the redo log. (That is, after a crash after CREATE SEQUENCE was committed, the effect of sequence operations will be observable fully or not at all.) When it comes to the durability of the updates of SEQUENCE in InnoDB, there is a clear analogy to MDEV-6076 Persistent AUTO_INCREMENT. The updates would be made persistent by the InnoDB redo log flush at transaction commit or rollback (or XA PREPARE), provided that innodb_log_flush_at_trx_commit=1. Similar to AUTO_INCREMENT, it is possible that the update of a SEQUENCE in a middle of transaction becomes durable before the COMMIT/ROLLBACK of the transaction, in case the InnoDB redo log is being flushed as a result of the a commit or rollback of some other transaction, or as a result of a redo log checkpoint that can be initiated at any time by operations that are writing redo log. dict_table_t::no_rollback(): Check if the table does not support rollback. BTR_NO_ROLLBACK: Logging and locking flags for no_rollback() tables. DICT_TF_BITS: Add the NO_ROLLBACK flag. row_ins_step(): Assign 0 to DB_ROW_ID and DB_TRX_ID, and skip any locking for no-rollback tables. There will be only a single row in no-rollback tables (or there must be a proper PRIMARY KEY). row_search_mvcc(): Execute the READ UNCOMMITTED code path for no-rollback tables. ha_innobase::external_lock(), ha_innobase::store_lock(): Block CREATE/DROP SEQUENCE in innodb_read_only mode. This probably has no effect for CREATE SEQUENCE, because already ha_innobase::create() should have been called (and refused) before external_lock() or store_lock() is called. ha_innobase::store_lock(): For CREATE SEQUENCE, do not acquire any InnoDB locks, even though TL_WRITE is being requested. (This is just a performance optimization.) innobase_copy_frm_flags_from_create_info(), row_drop_table_for_mysql(): Disable persistent statistics for no_rollback tables.
9 years ago
MDEV-10139 Support for InnoDB SEQUENCE objects We introduce a NO_ROLLBACK flag for InnoDB tables. This flag only works for tables that have a single index. Apart from undo logging, this flag will also prevent locking and the assignment of DB_ROW_ID or DB_TRX_ID, and imply READ UNCOMMITTED isolation. It is assumed that the SQL layer is guaranteeing mutual exclusion. After the initial insert of the single record during CREATE SEQUENCE, InnoDB will be updating the single record in-place. This is crash-safe thanks to the redo log. (That is, after a crash after CREATE SEQUENCE was committed, the effect of sequence operations will be observable fully or not at all.) When it comes to the durability of the updates of SEQUENCE in InnoDB, there is a clear analogy to MDEV-6076 Persistent AUTO_INCREMENT. The updates would be made persistent by the InnoDB redo log flush at transaction commit or rollback (or XA PREPARE), provided that innodb_log_flush_at_trx_commit=1. Similar to AUTO_INCREMENT, it is possible that the update of a SEQUENCE in a middle of transaction becomes durable before the COMMIT/ROLLBACK of the transaction, in case the InnoDB redo log is being flushed as a result of the a commit or rollback of some other transaction, or as a result of a redo log checkpoint that can be initiated at any time by operations that are writing redo log. dict_table_t::no_rollback(): Check if the table does not support rollback. BTR_NO_ROLLBACK: Logging and locking flags for no_rollback() tables. DICT_TF_BITS: Add the NO_ROLLBACK flag. row_ins_step(): Assign 0 to DB_ROW_ID and DB_TRX_ID, and skip any locking for no-rollback tables. There will be only a single row in no-rollback tables (or there must be a proper PRIMARY KEY). row_search_mvcc(): Execute the READ UNCOMMITTED code path for no-rollback tables. ha_innobase::external_lock(), ha_innobase::store_lock(): Block CREATE/DROP SEQUENCE in innodb_read_only mode. This probably has no effect for CREATE SEQUENCE, because already ha_innobase::create() should have been called (and refused) before external_lock() or store_lock() is called. ha_innobase::store_lock(): For CREATE SEQUENCE, do not acquire any InnoDB locks, even though TL_WRITE is being requested. (This is just a performance optimization.) innobase_copy_frm_flags_from_create_info(), row_drop_table_for_mysql(): Disable persistent statistics for no_rollback tables.
9 years ago
MDEV-12873 InnoDB SYS_TABLES.TYPE incompatibility for PAGE_COMPRESSED=YES in MariaDB 10.2.2 to 10.2.6 Remove the SHARED_SPACE flag that was erroneously introduced in MariaDB 10.2.2, and shift the SYS_TABLES.TYPE flags back to where they were before MariaDB 10.2.2. While doing this, ensure that tables created with affected MariaDB versions can be loaded, and also ensure that tables created with MySQL 5.7 using the TABLESPACE attribute cannot be loaded. MariaDB 10.2.2 picked the SHARED_SPACE flag from MySQL 5.7, shifting the MariaDB 10.1 flags PAGE_COMPRESSION, PAGE_COMPRESSION_LEVEL, ATOMIC_WRITES by one bit. The SHARED_SPACE flag would always be written as 0 by MariaDB, because MariaDB does not support CREATE TABLESPACE or CREATE TABLE...TABLESPACE for InnoDB. So, instead of the bits AALLLLCxxxxxxx we would have AALLLLC0xxxxxxx if the table was created with MariaDB 10.2.2 to 10.2.6. (AA=ATOMIC_WRITES, LLLL=PAGE_COMPRESSION_LEVEL, C=PAGE_COMPRESSED, xxxxxxx=7 bits that were not moved.) PAGE_COMPRESSED=NO implies LLLLC=00000. That is not a problem. If someone created a table in MariaDB 10.2.2 or 10.2.3 with the attribute ATOMIC_WRITES=OFF (value 2; AA=10) and without PAGE_COMPRESSED=YES or PAGE_COMPRESSION_LEVEL, the table should be rejected. We ignore this problem, because it should be unlikely for anyone to specify ATOMIC_WRITES=OFF, and because 10.2.2 and 10.2.2 were not mature releases. The value ATOMIC_WRITES=ON (1) would be interpreted as ATOMIC_WRITES=OFF, but starting with MariaDB 10.2.4 the ATOMIC_WRITES attribute is ignored. PAGE_COMPRESSED=YES implies that PAGE_COMPRESSION_LEVEL be between 1 and 9 and that ROW_FORMAT be COMPACT or DYNAMIC. Thus, the affected wrong bit pattern in SYS_TABLES.TYPE is of the form AALLLL10DB00001 where D signals the presence of a DATA DIRECTORY attribute and B is 1 for ROW_FORMAT=DYNAMIC and 0 for ROW_FORMAT=COMPACT. We must interpret this bit pattern as AALLLL1DB00001 (discarding the extraneous 0 bit). dict_sys_tables_rec_read(): Adjust the affected bit pattern when reading the SYS_TABLES.TYPE column. In case of invalid flags, report both SYS_TABLES.TYPE (after possible adjustment) and SYS_TABLES.MIX_LEN. dict_load_table_one(): Replace an unreachable condition on !dict_tf2_is_valid() with a debug assertion. The flags will already have been validated by dict_sys_tables_rec_read(); if that validation fails, dict_load_table_low() will have failed. fil_ibd_create(): Shorten an error message about a file pre-existing. Datafile::validate_to_dd(): Clarify an error message about tablespace flags mismatch. ha_innobase::open(): Remove an unnecessary warning message. dict_tf_is_valid(): Simplify and stricten the logic. Validate the values of PAGE_COMPRESSION. Remove error log output; let the callers handle that. DICT_TF_BITS: Remove ATOMIC_WRITES, PAGE_ENCRYPTION, PAGE_ENCRYPTION_KEY. The ATOMIC_WRITES is ignored once the SYS_TABLES.TYPE has been validated; there is no need to store it in dict_table_t::flags. The PAGE_ENCRYPTION and PAGE_ENCRYPTION_KEY are unused since MariaDB 10.1.4 (the GA release was 10.1.8). DICT_TF_BIT_MASK: Remove (unused). FSP_FLAGS_MEM_ATOMIC_WRITES: Remove (the flags are never read). row_import_read_v1(): Display an error if dict_tf_is_valid() fails.
8 years ago
MDEV-10139 Support for InnoDB SEQUENCE objects We introduce a NO_ROLLBACK flag for InnoDB tables. This flag only works for tables that have a single index. Apart from undo logging, this flag will also prevent locking and the assignment of DB_ROW_ID or DB_TRX_ID, and imply READ UNCOMMITTED isolation. It is assumed that the SQL layer is guaranteeing mutual exclusion. After the initial insert of the single record during CREATE SEQUENCE, InnoDB will be updating the single record in-place. This is crash-safe thanks to the redo log. (That is, after a crash after CREATE SEQUENCE was committed, the effect of sequence operations will be observable fully or not at all.) When it comes to the durability of the updates of SEQUENCE in InnoDB, there is a clear analogy to MDEV-6076 Persistent AUTO_INCREMENT. The updates would be made persistent by the InnoDB redo log flush at transaction commit or rollback (or XA PREPARE), provided that innodb_log_flush_at_trx_commit=1. Similar to AUTO_INCREMENT, it is possible that the update of a SEQUENCE in a middle of transaction becomes durable before the COMMIT/ROLLBACK of the transaction, in case the InnoDB redo log is being flushed as a result of the a commit or rollback of some other transaction, or as a result of a redo log checkpoint that can be initiated at any time by operations that are writing redo log. dict_table_t::no_rollback(): Check if the table does not support rollback. BTR_NO_ROLLBACK: Logging and locking flags for no_rollback() tables. DICT_TF_BITS: Add the NO_ROLLBACK flag. row_ins_step(): Assign 0 to DB_ROW_ID and DB_TRX_ID, and skip any locking for no-rollback tables. There will be only a single row in no-rollback tables (or there must be a proper PRIMARY KEY). row_search_mvcc(): Execute the READ UNCOMMITTED code path for no-rollback tables. ha_innobase::external_lock(), ha_innobase::store_lock(): Block CREATE/DROP SEQUENCE in innodb_read_only mode. This probably has no effect for CREATE SEQUENCE, because already ha_innobase::create() should have been called (and refused) before external_lock() or store_lock() is called. ha_innobase::store_lock(): For CREATE SEQUENCE, do not acquire any InnoDB locks, even though TL_WRITE is being requested. (This is just a performance optimization.) innobase_copy_frm_flags_from_create_info(), row_drop_table_for_mysql(): Disable persistent statistics for no_rollback tables.
9 years ago
MDEV-10139 Support for InnoDB SEQUENCE objects We introduce a NO_ROLLBACK flag for InnoDB tables. This flag only works for tables that have a single index. Apart from undo logging, this flag will also prevent locking and the assignment of DB_ROW_ID or DB_TRX_ID, and imply READ UNCOMMITTED isolation. It is assumed that the SQL layer is guaranteeing mutual exclusion. After the initial insert of the single record during CREATE SEQUENCE, InnoDB will be updating the single record in-place. This is crash-safe thanks to the redo log. (That is, after a crash after CREATE SEQUENCE was committed, the effect of sequence operations will be observable fully or not at all.) When it comes to the durability of the updates of SEQUENCE in InnoDB, there is a clear analogy to MDEV-6076 Persistent AUTO_INCREMENT. The updates would be made persistent by the InnoDB redo log flush at transaction commit or rollback (or XA PREPARE), provided that innodb_log_flush_at_trx_commit=1. Similar to AUTO_INCREMENT, it is possible that the update of a SEQUENCE in a middle of transaction becomes durable before the COMMIT/ROLLBACK of the transaction, in case the InnoDB redo log is being flushed as a result of the a commit or rollback of some other transaction, or as a result of a redo log checkpoint that can be initiated at any time by operations that are writing redo log. dict_table_t::no_rollback(): Check if the table does not support rollback. BTR_NO_ROLLBACK: Logging and locking flags for no_rollback() tables. DICT_TF_BITS: Add the NO_ROLLBACK flag. row_ins_step(): Assign 0 to DB_ROW_ID and DB_TRX_ID, and skip any locking for no-rollback tables. There will be only a single row in no-rollback tables (or there must be a proper PRIMARY KEY). row_search_mvcc(): Execute the READ UNCOMMITTED code path for no-rollback tables. ha_innobase::external_lock(), ha_innobase::store_lock(): Block CREATE/DROP SEQUENCE in innodb_read_only mode. This probably has no effect for CREATE SEQUENCE, because already ha_innobase::create() should have been called (and refused) before external_lock() or store_lock() is called. ha_innobase::store_lock(): For CREATE SEQUENCE, do not acquire any InnoDB locks, even though TL_WRITE is being requested. (This is just a performance optimization.) innobase_copy_frm_flags_from_create_info(), row_drop_table_for_mysql(): Disable persistent statistics for no_rollback tables.
9 years ago
9 years ago
9 years ago
9 years ago
9 years ago
12 years ago
MDEV-21907: InnoDB: Enable -Wconversion on clang and GCC The -Wconversion in GCC seems to be stricter than in clang. GCC at least since version 4.4.7 issues truncation warnings for assignments to bitfields, while clang 10 appears to only issue warnings when the sizes in bytes rounded to the nearest integer powers of 2 are different. Before GCC 10.0.0, -Wconversion required more casts and would not allow some operations, such as x<<=1 or x+=1 on a data type that is narrower than int. GCC 5 (but not GCC 4, GCC 6, or any later version) is complaining about x|=y even when x and y are compatible types that are narrower than int. Hence, we must rewrite some x|=y as x=static_cast<byte>(x|y) or similar, or we must disable -Wconversion. In GCC 6 and later, the warning for assigning wider to bitfields that are narrower than 8, 16, or 32 bits can be suppressed by applying a bitwise & with the exact bitmask of the bitfield. For older GCC, we must disable -Wconversion for GCC 4 or 5 in such cases. The bitwise negation operator appears to promote short integers to a wider type, and hence we must add explicit truncation casts around them. Microsoft Visual C does not allow a static_cast to truncate a constant, such as static_cast<byte>(1) truncating int. Hence, we will use the constructor-style cast byte(~1) for such cases. This has been tested at least with GCC 4.8.5, 5.4.0, 7.4.0, 9.2.1, 10.0.0, clang 9.0.1, 10.0.0, and MSVC 14.22.27905 (Microsoft Visual Studio 2019) on 64-bit and 32-bit targets (IA-32, AMD64, POWER 8, POWER 9, ARMv8).
6 years ago
9 years ago
11 years ago
11 years ago
MDEV-15662 Instant DROP COLUMN or changing the order of columns Allow ADD COLUMN anywhere in a table, not only adding as the last column. Allow instant DROP COLUMN and instant changing the order of columns. The added columns will always be added last in clustered index records. In new records, instantly dropped columns will be stored as NULL or empty when possible. Information about dropped and reordered columns will be written in a metadata BLOB (mblob), which is stored before the first 'user' field in the hidden metadata record at the start of the clustered index. The presence of mblob is indicated by setting the delete-mark flag in the metadata record. The metadata BLOB stores the number of clustered index fields, followed by an array of column information for each field. For dropped columns, we store the NOT NULL flag, the fixed length, and for variable-length columns, whether the maximum length exceeded 255 bytes. For non-dropped columns, we store the column position. Unlike with MDEV-11369, when a table becomes empty, it cannot be converted back to the canonical format. The reason for this is that other threads may hold cached objects such as row_prebuilt_t::ins_node that could refer to dropped or reordered index fields. For instant DROP COLUMN and ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC, we must store the n_core_null_bytes in the root page, so that the chain of node pointer records can be followed in order to reach the leftmost leaf page where the metadata record is located. If the mblob is present, we will zero-initialize the strings "infimum" and "supremum" in the root page, and use the last byte of "supremum" for storing the number of null bytes (which are allocated but useless on node pointer pages). This is necessary for btr_cur_instant_init_metadata() to be able to navigate to the mblob. If the PRIMARY KEY contains any variable-length column and some nullable columns were instantly dropped, the dict_index_t::n_nullable in the data dictionary could be smaller than it actually is in the non-leaf pages. Because of this, the non-leaf pages could use more bytes for the null flags than the data dictionary expects, and we could be reading the lengths of the variable-length columns from the wrong offset, and thus reading the child page number from wrong place. This is the result of two design mistakes that involve unnecessary storage of data: First, it is nonsense to store any data fields for the leftmost node pointer records, because the comparisons would be resolved by the MIN_REC_FLAG alone. Second, there cannot be any null fields in the clustered index node pointer fields, but we nevertheless reserve space for all the null flags. Limitations (future work): MDEV-17459 Allow instant ALTER TABLE even if FULLTEXT INDEX exists MDEV-17468 Avoid table rebuild on operations on generated columns MDEV-17494 Refuse ALGORITHM=INSTANT when the row size is too large btr_page_reorganize_low(): Preserve any metadata in the root page. Call lock_move_reorganize_page() only after restoring the "infimum" and "supremum" records, to avoid a memcmp() assertion failure. dict_col_t::DROPPED: Magic value for dict_col_t::ind. dict_col_t::clear_instant(): Renamed from dict_col_t::remove_instant(). Do not assert that the column was instantly added, because we sometimes call this unconditionally for all columns. Convert an instantly added column to a "core column". The old name remove_instant() could be mistaken to refer to "instant DROP COLUMN". dict_col_t::is_added(): Rename from dict_col_t::is_instant(). dtype_t::metadata_blob_init(): Initialize the mblob data type. dtuple_t::is_metadata(), dtuple_t::is_alter_metadata(), upd_t::is_metadata(), upd_t::is_alter_metadata(): Check if info_bits refer to a metadata record. dict_table_t::instant: Metadata about dropped or reordered columns. dict_table_t::prepare_instant(): Prepare ha_innobase_inplace_ctx::instant_table for instant ALTER TABLE. innobase_instant_try() will pass this to dict_table_t::instant_column(). On rollback, dict_table_t::rollback_instant() will be called. dict_table_t::instant_column(): Renamed from instant_add_column(). Add the parameter col_map so that columns can be reordered. Copy and adjust v_cols[] as well. dict_table_t::find(): Find an old column based on a new column number. dict_table_t::serialise_columns(), dict_table_t::deserialise_columns(): Convert the mblob. dict_index_t::instant_metadata(): Create the metadata record for instant ALTER TABLE. Invoke dict_table_t::serialise_columns(). dict_index_t::reconstruct_fields(): Invoked by dict_table_t::deserialise_columns(). dict_index_t::clear_instant_alter(): Move the fields for the dropped columns to the end, and sort the surviving index fields in ascending order of column position. ha_innobase::check_if_supported_inplace_alter(): Do not allow adding a FTS_DOC_ID column if a hidden FTS_DOC_ID column exists due to FULLTEXT INDEX. (This always required ALGORITHM=COPY.) instant_alter_column_possible(): Add a parameter for InnoDB table, to check for additional conditions, such as the maximum number of index fields. ha_innobase_inplace_ctx::first_alter_pos: The first column whose position is affected by instant ADD, DROP, or changing the order of columns. innobase_build_col_map(): Skip added virtual columns. prepare_inplace_add_virtual(): Correctly compute num_to_add_vcol. Remove some unnecessary code. Note that the call to innodb_base_col_setup() should be executed later. commit_try_norebuild(): If ctx->is_instant(), let the virtual columns be added or dropped by innobase_instant_try(). innobase_instant_try(): Fill in a zero default value for the hidden column FTS_DOC_ID (to reduce the work needed in MDEV-17459). If any columns were dropped or reordered (or added not last), delete any SYS_COLUMNS records for the following columns, and insert SYS_COLUMNS records for all subsequent stored columns as well as for all virtual columns. If any virtual column is dropped, rewrite all virtual column metadata. Use a shortcut only for adding virtual columns. This is because innobase_drop_virtual_try() assumes that the dropped virtual columns still exist in ctx->old_table. innodb_update_cols(): Renamed from innodb_update_n_cols(). innobase_add_one_virtual(), innobase_insert_sys_virtual(): Change the return type to bool, and invoke my_error() when detecting an error. innodb_insert_sys_columns(): Insert a record into SYS_COLUMNS. Refactored from innobase_add_one_virtual() and innobase_instant_add_col(). innobase_instant_add_col(): Replace the parameter dfield with type. innobase_instant_drop_cols(): Drop matching columns from SYS_COLUMNS and all columns from SYS_VIRTUAL. innobase_add_virtual_try(), innobase_drop_virtual_try(): Let the caller invoke innodb_update_cols(). innobase_rename_column_try(): Skip dropped columns. commit_cache_norebuild(): Update table->fts->doc_col. dict_mem_table_col_rename_low(): Skip dropped columns. trx_undo_rec_get_partial_row(): Skip dropped columns. trx_undo_update_rec_get_update(): Handle the metadata BLOB correctly. trx_undo_page_report_modify(): Avoid out-of-bounds access to record fields. Log metadata records consistently. Apparently, the first fields of a clustered index may be updated in an update_undo vector when the index is ID_IND of SYS_FOREIGN, as part of renaming the table during ALTER TABLE. Normally, updates of the PRIMARY KEY should be logged as delete-mark and an insert. row_undo_mod_parse_undo_rec(), row_purge_parse_undo_rec(): Use trx_undo_metadata. row_undo_mod_clust_low(): On metadata rollback, roll back the root page too. row_undo_mod_clust(): Relax an assertion. The delete-mark flag was repurposed for ALTER TABLE metadata records. row_rec_to_index_entry_impl(): Add the template parameter mblob and the optional parameter info_bits for specifying the desired new info bits. For the metadata tuple, allow conversion between the original format (ADD COLUMN only) and the generic format (with hidden BLOB). Add the optional parameter "pad" to determine whether the tuple should be padded to the index fields (on ALTER TABLE it should), or whether it should remain at its original size (on rollback). row_build_index_entry_low(): Clean up the code, removing redundant variables and conditions. For instantly dropped columns, generate a dummy value that is NULL, the empty string, or a fixed length of NUL bytes, depending on the type of the dropped column. row_upd_clust_rec_by_insert_inherit_func(): On the update of PRIMARY KEY of a record that contained a dropped column whose value was stored externally, we will be inserting a dummy NULL or empty string value to the field of the dropped column. The externally stored column would eventually be dropped when purge removes the delete-marked record for the old PRIMARY KEY value. btr_index_rec_validate(): Recognize the metadata record. btr_discard_only_page_on_level(): Preserve the generic instant ALTER TABLE metadata. btr_set_instant(): Replaces page_set_instant(). This sets a clustered index root page to the appropriate format, or upgrades from the MDEV-11369 instant ADD COLUMN to generic ALTER TABLE format. btr_cur_instant_init_low(): Read and validate the metadata BLOB page before reconstructing the dictionary information based on it. btr_cur_instant_init_metadata(): Do not read any lengths from the metadata record header before reading the BLOB. At this point, we would not actually know how many nullable fields the metadata record contains. btr_cur_instant_root_init(): Initialize n_core_null_bytes in one of two possible ways. btr_cur_trim(): Handle the mblob record. row_metadata_to_tuple(): Convert a metadata record to a data tuple, based on the new info_bits of the metadata record. btr_cur_pessimistic_update(): Invoke row_metadata_to_tuple() if needed. Invoke dtuple_convert_big_rec() for metadata records if the record is too large, or if the mblob is not yet marked as externally stored. btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete(): When the last user record is deleted, do not delete the generic instant ALTER TABLE metadata record. Only delete MDEV-11369 instant ADD COLUMN metadata records. btr_cur_optimistic_insert(): Avoid unnecessary computation of rec_size. btr_pcur_store_position(): Allow a logically empty page to contain a metadata record for generic ALTER TABLE. REC_INFO_DEFAULT_ROW_ADD: Renamed from REC_INFO_DEFAULT_ROW. This is for the old instant ADD COLUMN (MDEV-11369) only. REC_INFO_DEFAULT_ROW_ALTER: The more generic metadata record, with additional information for dropped or reordered columns. rec_info_bits_valid(): Remove. The only case when this would fail is when the record is the generic ALTER TABLE metadata record. rec_is_alter_metadata(): Check if a record is the metadata record for instant ALTER TABLE (other than ADD COLUMN). NOTE: This function must not be invoked on node pointer records, because the delete-mark flag in those records may be set (it is garbage), and then a debug assertion could fail because index->is_instant() does not necessarily hold. rec_is_add_metadata(): Check if a record is MDEV-11369 ADD COLUMN metadata record (not more generic instant ALTER TABLE). rec_get_converted_size_comp_prefix_low(): Assume that the metadata field will be stored externally. In dtuple_convert_big_rec() during the rec_get_converted_size() call, it would not be there yet. rec_get_converted_size_comp(): Replace status,fields,n_fields with tuple. rec_init_offsets_comp_ordinary(), rec_get_converted_size_comp_prefix_low(), rec_convert_dtuple_to_rec_comp(): Add template<bool mblob = false>. With mblob=true, process a record with a metadata BLOB. rec_copy_prefix_to_buf(): Assert that no fields beyond the key and system columns are being copied. Exclude the metadata BLOB field. rec_convert_dtuple_to_metadata_comp(): Convert an alter metadata tuple into a record. row_upd_index_replace_metadata(): Apply an update vector to an alter_metadata tuple. row_log_allocate(): Replace dict_index_t::is_instant() with a more appropriate condition that ignores dict_table_t::instant. Only a table on which the MDEV-11369 ADD COLUMN was performed can "lose its instantness" when it becomes empty. After instant DROP COLUMN or reordering columns, we cannot simply convert the table to the canonical format, because the data dictionary cache and all possibly existing references to it from other client connection threads would have to be adjusted. row_quiesce_write_index_fields(): Do not crash when the table contains an instantly dropped column. Thanks to Thirunarayanan Balathandayuthapani for discussing the design and implementing an initial prototype of this. Thanks to Matthias Leich for testing.
7 years ago
MDEV-11369 Instant ADD COLUMN for InnoDB For InnoDB tables, adding, dropping and reordering columns has required a rebuild of the table and all its indexes. Since MySQL 5.6 (and MariaDB 10.0) this has been supported online (LOCK=NONE), allowing concurrent modification of the tables. This work revises the InnoDB ROW_FORMAT=REDUNDANT, ROW_FORMAT=COMPACT and ROW_FORMAT=DYNAMIC so that columns can be appended instantaneously, with only minor changes performed to the table structure. The counter innodb_instant_alter_column in INFORMATION_SCHEMA.GLOBAL_STATUS is incremented whenever a table rebuild operation is converted into an instant ADD COLUMN operation. ROW_FORMAT=COMPRESSED tables will not support instant ADD COLUMN. Some usability limitations will be addressed in subsequent work: MDEV-13134 Introduce ALTER TABLE attributes ALGORITHM=NOCOPY and ALGORITHM=INSTANT MDEV-14016 Allow instant ADD COLUMN, ADD INDEX, LOCK=NONE The format of the clustered index (PRIMARY KEY) is changed as follows: (1) The FIL_PAGE_TYPE of the root page will be FIL_PAGE_TYPE_INSTANT, and a new field PAGE_INSTANT will contain the original number of fields in the clustered index ('core' fields). If instant ADD COLUMN has not been used or the table becomes empty, or the very first instant ADD COLUMN operation is rolled back, the fields PAGE_INSTANT and FIL_PAGE_TYPE will be reset to 0 and FIL_PAGE_INDEX. (2) A special 'default row' record is inserted into the leftmost leaf, between the page infimum and the first user record. This record is distinguished by the REC_INFO_MIN_REC_FLAG, and it is otherwise in the same format as records that contain values for the instantly added columns. This 'default row' always has the same number of fields as the clustered index according to the table definition. The values of 'core' fields are to be ignored. For other fields, the 'default row' will contain the default values as they were during the ALTER TABLE statement. (If the column default values are changed later, those values will only be stored in the .frm file. The 'default row' will contain the original evaluated values, which must be the same for every row.) The 'default row' must be completely hidden from higher-level access routines. Assertions have been added to ensure that no 'default row' is ever present in the adaptive hash index or in locked records. The 'default row' is never delete-marked. (3) In clustered index leaf page records, the number of fields must reside between the number of 'core' fields (dict_index_t::n_core_fields introduced in this work) and dict_index_t::n_fields. If the number of fields is less than dict_index_t::n_fields, the missing fields are replaced with the column value of the 'default row'. Note: The number of fields in the record may shrink if some of the last instantly added columns are updated to the value that is in the 'default row'. The function btr_cur_trim() implements this 'compression' on update and rollback; dtuple::trim() implements it on insert. (4) In ROW_FORMAT=COMPACT and ROW_FORMAT=DYNAMIC records, the new status value REC_STATUS_COLUMNS_ADDED will indicate the presence of a new record header that will encode n_fields-n_core_fields-1 in 1 or 2 bytes. (In ROW_FORMAT=REDUNDANT records, the record header always explicitly encodes the number of fields.) We introduce the undo log record type TRX_UNDO_INSERT_DEFAULT for covering the insert of the 'default row' record when instant ADD COLUMN is used for the first time. Subsequent instant ADD COLUMN can use TRX_UNDO_UPD_EXIST_REC. This is joint work with Vin Chen (陈福荣) from Tencent. The design that was discussed in April 2017 would not have allowed import or export of data files, because instead of the 'default row' it would have introduced a data dictionary table. The test rpl.rpl_alter_instant is exactly as contributed in pull request #408. The test innodb.instant_alter is based on a contributed test. The redo log record format changes for ROW_FORMAT=DYNAMIC and ROW_FORMAT=COMPACT are as contributed. (With this change present, crash recovery from MariaDB 10.3.1 will fail in spectacular ways!) Also the semantics of higher-level redo log records that modify the PAGE_INSTANT field is changed. The redo log format version identifier was already changed to LOG_HEADER_FORMAT_CURRENT=103 in MariaDB 10.3.1. Everything else has been rewritten by me. Thanks to Elena Stepanova, the code has been tested extensively. When rolling back an instant ADD COLUMN operation, we must empty the PAGE_FREE list after deleting or shortening the 'default row' record, by calling either btr_page_empty() or btr_page_reorganize(). We must know the size of each entry in the PAGE_FREE list. If rollback left a freed copy of the 'default row' in the PAGE_FREE list, we would be unable to determine its size (if it is in ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC) because it would contain more fields than the rolled-back definition of the clustered index. UNIV_SQL_DEFAULT: A new special constant that designates an instantly added column that is not present in the clustered index record. len_is_stored(): Check if a length is an actual length. There are two magic length values: UNIV_SQL_DEFAULT, UNIV_SQL_NULL. dict_col_t::def_val: The 'default row' value of the column. If the column is not added instantly, def_val.len will be UNIV_SQL_DEFAULT. dict_col_t: Add the accessors is_virtual(), is_nullable(), is_instant(), instant_value(). dict_col_t::remove_instant(): Remove the 'instant ADD' status of a column. dict_col_t::name(const dict_table_t& table): Replaces dict_table_get_col_name(). dict_index_t::n_core_fields: The original number of fields. For secondary indexes and if instant ADD COLUMN has not been used, this will be equal to dict_index_t::n_fields. dict_index_t::n_core_null_bytes: Number of bytes needed to represent the null flags; usually equal to UT_BITS_IN_BYTES(n_nullable). dict_index_t::NO_CORE_NULL_BYTES: Magic value signalling that n_core_null_bytes was not initialized yet from the clustered index root page. dict_index_t: Add the accessors is_instant(), is_clust(), get_n_nullable(), instant_field_value(). dict_index_t::instant_add_field(): Adjust clustered index metadata for instant ADD COLUMN. dict_index_t::remove_instant(): Remove the 'instant ADD' status of a clustered index when the table becomes empty, or the very first instant ADD COLUMN operation is rolled back. dict_table_t: Add the accessors is_instant(), is_temporary(), supports_instant(). dict_table_t::instant_add_column(): Adjust metadata for instant ADD COLUMN. dict_table_t::rollback_instant(): Adjust metadata on the rollback of instant ADD COLUMN. prepare_inplace_alter_table_dict(): First create the ctx->new_table, and only then decide if the table really needs to be rebuilt. We must split the creation of table or index metadata from the creation of the dictionary table records and the creation of the data. In this way, we can transform a table-rebuilding operation into an instant ADD COLUMN operation. Dictionary objects will only be added to cache when table rebuilding or index creation is needed. The ctx->instant_table will never be added to cache. dict_table_t::add_to_cache(): Modified and renamed from dict_table_add_to_cache(). Do not modify the table metadata. Let the callers invoke dict_table_add_system_columns() and if needed, set can_be_evicted. dict_create_sys_tables_tuple(), dict_create_table_step(): Omit the system columns (which will now exist in the dict_table_t object already at this point). dict_create_table_step(): Expect the callers to invoke dict_table_add_system_columns(). pars_create_table(): Before creating the table creation execution graph, invoke dict_table_add_system_columns(). row_create_table_for_mysql(): Expect all callers to invoke dict_table_add_system_columns(). create_index_dict(): Replaces row_merge_create_index_graph(). innodb_update_n_cols(): Renamed from innobase_update_n_virtual(). Call my_error() if an error occurs. btr_cur_instant_init(), btr_cur_instant_init_low(), btr_cur_instant_root_init(): Load additional metadata from the clustered index and set dict_index_t::n_core_null_bytes. This is invoked when table metadata is first loaded into the data dictionary. dict_boot(): Initialize n_core_null_bytes for the four hard-coded dictionary tables. dict_create_index_step(): Initialize n_core_null_bytes. This is executed as part of CREATE TABLE. dict_index_build_internal_clust(): Initialize n_core_null_bytes to NO_CORE_NULL_BYTES if table->supports_instant(). row_create_index_for_mysql(): Initialize n_core_null_bytes for CREATE TEMPORARY TABLE. commit_cache_norebuild(): Call the code to rename or enlarge columns in the cache only if instant ADD COLUMN is not being used. (Instant ADD COLUMN would copy all column metadata from instant_table to old_table, including the names and lengths.) PAGE_INSTANT: A new 13-bit field for storing dict_index_t::n_core_fields. This is repurposing the 16-bit field PAGE_DIRECTION, of which only the least significant 3 bits were used. The original byte containing PAGE_DIRECTION will be accessible via the new constant PAGE_DIRECTION_B. page_get_instant(), page_set_instant(): Accessors for the PAGE_INSTANT. page_ptr_get_direction(), page_get_direction(), page_ptr_set_direction(): Accessors for PAGE_DIRECTION. page_direction_reset(): Reset PAGE_DIRECTION, PAGE_N_DIRECTION. page_direction_increment(): Increment PAGE_N_DIRECTION and set PAGE_DIRECTION. rec_get_offsets(): Use the 'leaf' parameter for non-debug purposes, and assume that heap_no is always set. Initialize all dict_index_t::n_fields for ROW_FORMAT=REDUNDANT records, even if the record contains fewer fields. rec_offs_make_valid(): Add the parameter 'leaf'. rec_copy_prefix_to_dtuple(): Assert that the tuple is only built on the core fields. Instant ADD COLUMN only applies to the clustered index, and we should never build a search key that has more than the PRIMARY KEY and possibly DB_TRX_ID,DB_ROLL_PTR. All these columns are always present. dict_index_build_data_tuple(): Remove assertions that would be duplicated in rec_copy_prefix_to_dtuple(). rec_init_offsets(): Support ROW_FORMAT=REDUNDANT records whose number of fields is between n_core_fields and n_fields. cmp_rec_rec_with_match(): Implement the comparison between two MIN_REC_FLAG records. trx_t::in_rollback: Make the field available in non-debug builds. trx_start_for_ddl_low(): Remove dangerous error-tolerance. A dictionary transaction must be flagged as such before it has generated any undo log records. This is because trx_undo_assign_undo() will mark the transaction as a dictionary transaction in the undo log header right before the very first undo log record is being written. btr_index_rec_validate(): Account for instant ADD COLUMN row_undo_ins_remove_clust_rec(): On the rollback of an insert into SYS_COLUMNS, revert instant ADD COLUMN in the cache by removing the last column from the table and the clustered index. row_search_on_row_ref(), row_undo_mod_parse_undo_rec(), row_undo_mod(), trx_undo_update_rec_get_update(): Handle the 'default row' as a special case. dtuple_t::trim(index): Omit a redundant suffix of an index tuple right before insert or update. After instant ADD COLUMN, if the last fields of a clustered index tuple match the 'default row', there is no need to store them. While trimming the entry, we must hold a page latch, so that the table cannot be emptied and the 'default row' be deleted. btr_cur_optimistic_update(), btr_cur_pessimistic_update(), row_upd_clust_rec_by_insert(), row_ins_clust_index_entry_low(): Invoke dtuple_t::trim() if needed. row_ins_clust_index_entry(): Restore dtuple_t::n_fields after calling row_ins_clust_index_entry_low(). rec_get_converted_size(), rec_get_converted_size_comp(): Allow the number of fields to be between n_core_fields and n_fields. Do not support infimum,supremum. They are never supposed to be stored in dtuple_t, because page creation nowadays uses a lower-level method for initializing them. rec_convert_dtuple_to_rec_comp(): Assign the status bits based on the number of fields. btr_cur_trim(): In an update, trim the index entry as needed. For the 'default row', handle rollback specially. For user records, omit fields that match the 'default row'. btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete(): Skip locking and adaptive hash index for the 'default row'. row_log_table_apply_convert_mrec(): Replace 'default row' values if needed. In the temporary file that is applied by row_log_table_apply(), we must identify whether the records contain the extra header for instantly added columns. For now, we will allocate an additional byte for this for ROW_T_INSERT and ROW_T_UPDATE records when the source table has been subject to instant ADD COLUMN. The ROW_T_DELETE records are fine, as they will be converted and will only contain 'core' columns (PRIMARY KEY and some system columns) that are converted from dtuple_t. rec_get_converted_size_temp(), rec_init_offsets_temp(), rec_convert_dtuple_to_temp(): Add the parameter 'status'. REC_INFO_DEFAULT_ROW = REC_INFO_MIN_REC_FLAG | REC_STATUS_COLUMNS_ADDED: An info_bits constant for distinguishing the 'default row' record. rec_comp_status_t: An enum of the status bit values. rec_leaf_format: An enum that replaces the bool parameter of rec_init_offsets_comp_ordinary().
8 years ago
MDEV-15662 Instant DROP COLUMN or changing the order of columns Allow ADD COLUMN anywhere in a table, not only adding as the last column. Allow instant DROP COLUMN and instant changing the order of columns. The added columns will always be added last in clustered index records. In new records, instantly dropped columns will be stored as NULL or empty when possible. Information about dropped and reordered columns will be written in a metadata BLOB (mblob), which is stored before the first 'user' field in the hidden metadata record at the start of the clustered index. The presence of mblob is indicated by setting the delete-mark flag in the metadata record. The metadata BLOB stores the number of clustered index fields, followed by an array of column information for each field. For dropped columns, we store the NOT NULL flag, the fixed length, and for variable-length columns, whether the maximum length exceeded 255 bytes. For non-dropped columns, we store the column position. Unlike with MDEV-11369, when a table becomes empty, it cannot be converted back to the canonical format. The reason for this is that other threads may hold cached objects such as row_prebuilt_t::ins_node that could refer to dropped or reordered index fields. For instant DROP COLUMN and ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC, we must store the n_core_null_bytes in the root page, so that the chain of node pointer records can be followed in order to reach the leftmost leaf page where the metadata record is located. If the mblob is present, we will zero-initialize the strings "infimum" and "supremum" in the root page, and use the last byte of "supremum" for storing the number of null bytes (which are allocated but useless on node pointer pages). This is necessary for btr_cur_instant_init_metadata() to be able to navigate to the mblob. If the PRIMARY KEY contains any variable-length column and some nullable columns were instantly dropped, the dict_index_t::n_nullable in the data dictionary could be smaller than it actually is in the non-leaf pages. Because of this, the non-leaf pages could use more bytes for the null flags than the data dictionary expects, and we could be reading the lengths of the variable-length columns from the wrong offset, and thus reading the child page number from wrong place. This is the result of two design mistakes that involve unnecessary storage of data: First, it is nonsense to store any data fields for the leftmost node pointer records, because the comparisons would be resolved by the MIN_REC_FLAG alone. Second, there cannot be any null fields in the clustered index node pointer fields, but we nevertheless reserve space for all the null flags. Limitations (future work): MDEV-17459 Allow instant ALTER TABLE even if FULLTEXT INDEX exists MDEV-17468 Avoid table rebuild on operations on generated columns MDEV-17494 Refuse ALGORITHM=INSTANT when the row size is too large btr_page_reorganize_low(): Preserve any metadata in the root page. Call lock_move_reorganize_page() only after restoring the "infimum" and "supremum" records, to avoid a memcmp() assertion failure. dict_col_t::DROPPED: Magic value for dict_col_t::ind. dict_col_t::clear_instant(): Renamed from dict_col_t::remove_instant(). Do not assert that the column was instantly added, because we sometimes call this unconditionally for all columns. Convert an instantly added column to a "core column". The old name remove_instant() could be mistaken to refer to "instant DROP COLUMN". dict_col_t::is_added(): Rename from dict_col_t::is_instant(). dtype_t::metadata_blob_init(): Initialize the mblob data type. dtuple_t::is_metadata(), dtuple_t::is_alter_metadata(), upd_t::is_metadata(), upd_t::is_alter_metadata(): Check if info_bits refer to a metadata record. dict_table_t::instant: Metadata about dropped or reordered columns. dict_table_t::prepare_instant(): Prepare ha_innobase_inplace_ctx::instant_table for instant ALTER TABLE. innobase_instant_try() will pass this to dict_table_t::instant_column(). On rollback, dict_table_t::rollback_instant() will be called. dict_table_t::instant_column(): Renamed from instant_add_column(). Add the parameter col_map so that columns can be reordered. Copy and adjust v_cols[] as well. dict_table_t::find(): Find an old column based on a new column number. dict_table_t::serialise_columns(), dict_table_t::deserialise_columns(): Convert the mblob. dict_index_t::instant_metadata(): Create the metadata record for instant ALTER TABLE. Invoke dict_table_t::serialise_columns(). dict_index_t::reconstruct_fields(): Invoked by dict_table_t::deserialise_columns(). dict_index_t::clear_instant_alter(): Move the fields for the dropped columns to the end, and sort the surviving index fields in ascending order of column position. ha_innobase::check_if_supported_inplace_alter(): Do not allow adding a FTS_DOC_ID column if a hidden FTS_DOC_ID column exists due to FULLTEXT INDEX. (This always required ALGORITHM=COPY.) instant_alter_column_possible(): Add a parameter for InnoDB table, to check for additional conditions, such as the maximum number of index fields. ha_innobase_inplace_ctx::first_alter_pos: The first column whose position is affected by instant ADD, DROP, or changing the order of columns. innobase_build_col_map(): Skip added virtual columns. prepare_inplace_add_virtual(): Correctly compute num_to_add_vcol. Remove some unnecessary code. Note that the call to innodb_base_col_setup() should be executed later. commit_try_norebuild(): If ctx->is_instant(), let the virtual columns be added or dropped by innobase_instant_try(). innobase_instant_try(): Fill in a zero default value for the hidden column FTS_DOC_ID (to reduce the work needed in MDEV-17459). If any columns were dropped or reordered (or added not last), delete any SYS_COLUMNS records for the following columns, and insert SYS_COLUMNS records for all subsequent stored columns as well as for all virtual columns. If any virtual column is dropped, rewrite all virtual column metadata. Use a shortcut only for adding virtual columns. This is because innobase_drop_virtual_try() assumes that the dropped virtual columns still exist in ctx->old_table. innodb_update_cols(): Renamed from innodb_update_n_cols(). innobase_add_one_virtual(), innobase_insert_sys_virtual(): Change the return type to bool, and invoke my_error() when detecting an error. innodb_insert_sys_columns(): Insert a record into SYS_COLUMNS. Refactored from innobase_add_one_virtual() and innobase_instant_add_col(). innobase_instant_add_col(): Replace the parameter dfield with type. innobase_instant_drop_cols(): Drop matching columns from SYS_COLUMNS and all columns from SYS_VIRTUAL. innobase_add_virtual_try(), innobase_drop_virtual_try(): Let the caller invoke innodb_update_cols(). innobase_rename_column_try(): Skip dropped columns. commit_cache_norebuild(): Update table->fts->doc_col. dict_mem_table_col_rename_low(): Skip dropped columns. trx_undo_rec_get_partial_row(): Skip dropped columns. trx_undo_update_rec_get_update(): Handle the metadata BLOB correctly. trx_undo_page_report_modify(): Avoid out-of-bounds access to record fields. Log metadata records consistently. Apparently, the first fields of a clustered index may be updated in an update_undo vector when the index is ID_IND of SYS_FOREIGN, as part of renaming the table during ALTER TABLE. Normally, updates of the PRIMARY KEY should be logged as delete-mark and an insert. row_undo_mod_parse_undo_rec(), row_purge_parse_undo_rec(): Use trx_undo_metadata. row_undo_mod_clust_low(): On metadata rollback, roll back the root page too. row_undo_mod_clust(): Relax an assertion. The delete-mark flag was repurposed for ALTER TABLE metadata records. row_rec_to_index_entry_impl(): Add the template parameter mblob and the optional parameter info_bits for specifying the desired new info bits. For the metadata tuple, allow conversion between the original format (ADD COLUMN only) and the generic format (with hidden BLOB). Add the optional parameter "pad" to determine whether the tuple should be padded to the index fields (on ALTER TABLE it should), or whether it should remain at its original size (on rollback). row_build_index_entry_low(): Clean up the code, removing redundant variables and conditions. For instantly dropped columns, generate a dummy value that is NULL, the empty string, or a fixed length of NUL bytes, depending on the type of the dropped column. row_upd_clust_rec_by_insert_inherit_func(): On the update of PRIMARY KEY of a record that contained a dropped column whose value was stored externally, we will be inserting a dummy NULL or empty string value to the field of the dropped column. The externally stored column would eventually be dropped when purge removes the delete-marked record for the old PRIMARY KEY value. btr_index_rec_validate(): Recognize the metadata record. btr_discard_only_page_on_level(): Preserve the generic instant ALTER TABLE metadata. btr_set_instant(): Replaces page_set_instant(). This sets a clustered index root page to the appropriate format, or upgrades from the MDEV-11369 instant ADD COLUMN to generic ALTER TABLE format. btr_cur_instant_init_low(): Read and validate the metadata BLOB page before reconstructing the dictionary information based on it. btr_cur_instant_init_metadata(): Do not read any lengths from the metadata record header before reading the BLOB. At this point, we would not actually know how many nullable fields the metadata record contains. btr_cur_instant_root_init(): Initialize n_core_null_bytes in one of two possible ways. btr_cur_trim(): Handle the mblob record. row_metadata_to_tuple(): Convert a metadata record to a data tuple, based on the new info_bits of the metadata record. btr_cur_pessimistic_update(): Invoke row_metadata_to_tuple() if needed. Invoke dtuple_convert_big_rec() for metadata records if the record is too large, or if the mblob is not yet marked as externally stored. btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete(): When the last user record is deleted, do not delete the generic instant ALTER TABLE metadata record. Only delete MDEV-11369 instant ADD COLUMN metadata records. btr_cur_optimistic_insert(): Avoid unnecessary computation of rec_size. btr_pcur_store_position(): Allow a logically empty page to contain a metadata record for generic ALTER TABLE. REC_INFO_DEFAULT_ROW_ADD: Renamed from REC_INFO_DEFAULT_ROW. This is for the old instant ADD COLUMN (MDEV-11369) only. REC_INFO_DEFAULT_ROW_ALTER: The more generic metadata record, with additional information for dropped or reordered columns. rec_info_bits_valid(): Remove. The only case when this would fail is when the record is the generic ALTER TABLE metadata record. rec_is_alter_metadata(): Check if a record is the metadata record for instant ALTER TABLE (other than ADD COLUMN). NOTE: This function must not be invoked on node pointer records, because the delete-mark flag in those records may be set (it is garbage), and then a debug assertion could fail because index->is_instant() does not necessarily hold. rec_is_add_metadata(): Check if a record is MDEV-11369 ADD COLUMN metadata record (not more generic instant ALTER TABLE). rec_get_converted_size_comp_prefix_low(): Assume that the metadata field will be stored externally. In dtuple_convert_big_rec() during the rec_get_converted_size() call, it would not be there yet. rec_get_converted_size_comp(): Replace status,fields,n_fields with tuple. rec_init_offsets_comp_ordinary(), rec_get_converted_size_comp_prefix_low(), rec_convert_dtuple_to_rec_comp(): Add template<bool mblob = false>. With mblob=true, process a record with a metadata BLOB. rec_copy_prefix_to_buf(): Assert that no fields beyond the key and system columns are being copied. Exclude the metadata BLOB field. rec_convert_dtuple_to_metadata_comp(): Convert an alter metadata tuple into a record. row_upd_index_replace_metadata(): Apply an update vector to an alter_metadata tuple. row_log_allocate(): Replace dict_index_t::is_instant() with a more appropriate condition that ignores dict_table_t::instant. Only a table on which the MDEV-11369 ADD COLUMN was performed can "lose its instantness" when it becomes empty. After instant DROP COLUMN or reordering columns, we cannot simply convert the table to the canonical format, because the data dictionary cache and all possibly existing references to it from other client connection threads would have to be adjusted. row_quiesce_write_index_fields(): Do not crash when the table contains an instantly dropped column. Thanks to Thirunarayanan Balathandayuthapani for discussing the design and implementing an initial prototype of this. Thanks to Matthias Leich for testing.
7 years ago
MDEV-11369 Instant ADD COLUMN for InnoDB For InnoDB tables, adding, dropping and reordering columns has required a rebuild of the table and all its indexes. Since MySQL 5.6 (and MariaDB 10.0) this has been supported online (LOCK=NONE), allowing concurrent modification of the tables. This work revises the InnoDB ROW_FORMAT=REDUNDANT, ROW_FORMAT=COMPACT and ROW_FORMAT=DYNAMIC so that columns can be appended instantaneously, with only minor changes performed to the table structure. The counter innodb_instant_alter_column in INFORMATION_SCHEMA.GLOBAL_STATUS is incremented whenever a table rebuild operation is converted into an instant ADD COLUMN operation. ROW_FORMAT=COMPRESSED tables will not support instant ADD COLUMN. Some usability limitations will be addressed in subsequent work: MDEV-13134 Introduce ALTER TABLE attributes ALGORITHM=NOCOPY and ALGORITHM=INSTANT MDEV-14016 Allow instant ADD COLUMN, ADD INDEX, LOCK=NONE The format of the clustered index (PRIMARY KEY) is changed as follows: (1) The FIL_PAGE_TYPE of the root page will be FIL_PAGE_TYPE_INSTANT, and a new field PAGE_INSTANT will contain the original number of fields in the clustered index ('core' fields). If instant ADD COLUMN has not been used or the table becomes empty, or the very first instant ADD COLUMN operation is rolled back, the fields PAGE_INSTANT and FIL_PAGE_TYPE will be reset to 0 and FIL_PAGE_INDEX. (2) A special 'default row' record is inserted into the leftmost leaf, between the page infimum and the first user record. This record is distinguished by the REC_INFO_MIN_REC_FLAG, and it is otherwise in the same format as records that contain values for the instantly added columns. This 'default row' always has the same number of fields as the clustered index according to the table definition. The values of 'core' fields are to be ignored. For other fields, the 'default row' will contain the default values as they were during the ALTER TABLE statement. (If the column default values are changed later, those values will only be stored in the .frm file. The 'default row' will contain the original evaluated values, which must be the same for every row.) The 'default row' must be completely hidden from higher-level access routines. Assertions have been added to ensure that no 'default row' is ever present in the adaptive hash index or in locked records. The 'default row' is never delete-marked. (3) In clustered index leaf page records, the number of fields must reside between the number of 'core' fields (dict_index_t::n_core_fields introduced in this work) and dict_index_t::n_fields. If the number of fields is less than dict_index_t::n_fields, the missing fields are replaced with the column value of the 'default row'. Note: The number of fields in the record may shrink if some of the last instantly added columns are updated to the value that is in the 'default row'. The function btr_cur_trim() implements this 'compression' on update and rollback; dtuple::trim() implements it on insert. (4) In ROW_FORMAT=COMPACT and ROW_FORMAT=DYNAMIC records, the new status value REC_STATUS_COLUMNS_ADDED will indicate the presence of a new record header that will encode n_fields-n_core_fields-1 in 1 or 2 bytes. (In ROW_FORMAT=REDUNDANT records, the record header always explicitly encodes the number of fields.) We introduce the undo log record type TRX_UNDO_INSERT_DEFAULT for covering the insert of the 'default row' record when instant ADD COLUMN is used for the first time. Subsequent instant ADD COLUMN can use TRX_UNDO_UPD_EXIST_REC. This is joint work with Vin Chen (陈福荣) from Tencent. The design that was discussed in April 2017 would not have allowed import or export of data files, because instead of the 'default row' it would have introduced a data dictionary table. The test rpl.rpl_alter_instant is exactly as contributed in pull request #408. The test innodb.instant_alter is based on a contributed test. The redo log record format changes for ROW_FORMAT=DYNAMIC and ROW_FORMAT=COMPACT are as contributed. (With this change present, crash recovery from MariaDB 10.3.1 will fail in spectacular ways!) Also the semantics of higher-level redo log records that modify the PAGE_INSTANT field is changed. The redo log format version identifier was already changed to LOG_HEADER_FORMAT_CURRENT=103 in MariaDB 10.3.1. Everything else has been rewritten by me. Thanks to Elena Stepanova, the code has been tested extensively. When rolling back an instant ADD COLUMN operation, we must empty the PAGE_FREE list after deleting or shortening the 'default row' record, by calling either btr_page_empty() or btr_page_reorganize(). We must know the size of each entry in the PAGE_FREE list. If rollback left a freed copy of the 'default row' in the PAGE_FREE list, we would be unable to determine its size (if it is in ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC) because it would contain more fields than the rolled-back definition of the clustered index. UNIV_SQL_DEFAULT: A new special constant that designates an instantly added column that is not present in the clustered index record. len_is_stored(): Check if a length is an actual length. There are two magic length values: UNIV_SQL_DEFAULT, UNIV_SQL_NULL. dict_col_t::def_val: The 'default row' value of the column. If the column is not added instantly, def_val.len will be UNIV_SQL_DEFAULT. dict_col_t: Add the accessors is_virtual(), is_nullable(), is_instant(), instant_value(). dict_col_t::remove_instant(): Remove the 'instant ADD' status of a column. dict_col_t::name(const dict_table_t& table): Replaces dict_table_get_col_name(). dict_index_t::n_core_fields: The original number of fields. For secondary indexes and if instant ADD COLUMN has not been used, this will be equal to dict_index_t::n_fields. dict_index_t::n_core_null_bytes: Number of bytes needed to represent the null flags; usually equal to UT_BITS_IN_BYTES(n_nullable). dict_index_t::NO_CORE_NULL_BYTES: Magic value signalling that n_core_null_bytes was not initialized yet from the clustered index root page. dict_index_t: Add the accessors is_instant(), is_clust(), get_n_nullable(), instant_field_value(). dict_index_t::instant_add_field(): Adjust clustered index metadata for instant ADD COLUMN. dict_index_t::remove_instant(): Remove the 'instant ADD' status of a clustered index when the table becomes empty, or the very first instant ADD COLUMN operation is rolled back. dict_table_t: Add the accessors is_instant(), is_temporary(), supports_instant(). dict_table_t::instant_add_column(): Adjust metadata for instant ADD COLUMN. dict_table_t::rollback_instant(): Adjust metadata on the rollback of instant ADD COLUMN. prepare_inplace_alter_table_dict(): First create the ctx->new_table, and only then decide if the table really needs to be rebuilt. We must split the creation of table or index metadata from the creation of the dictionary table records and the creation of the data. In this way, we can transform a table-rebuilding operation into an instant ADD COLUMN operation. Dictionary objects will only be added to cache when table rebuilding or index creation is needed. The ctx->instant_table will never be added to cache. dict_table_t::add_to_cache(): Modified and renamed from dict_table_add_to_cache(). Do not modify the table metadata. Let the callers invoke dict_table_add_system_columns() and if needed, set can_be_evicted. dict_create_sys_tables_tuple(), dict_create_table_step(): Omit the system columns (which will now exist in the dict_table_t object already at this point). dict_create_table_step(): Expect the callers to invoke dict_table_add_system_columns(). pars_create_table(): Before creating the table creation execution graph, invoke dict_table_add_system_columns(). row_create_table_for_mysql(): Expect all callers to invoke dict_table_add_system_columns(). create_index_dict(): Replaces row_merge_create_index_graph(). innodb_update_n_cols(): Renamed from innobase_update_n_virtual(). Call my_error() if an error occurs. btr_cur_instant_init(), btr_cur_instant_init_low(), btr_cur_instant_root_init(): Load additional metadata from the clustered index and set dict_index_t::n_core_null_bytes. This is invoked when table metadata is first loaded into the data dictionary. dict_boot(): Initialize n_core_null_bytes for the four hard-coded dictionary tables. dict_create_index_step(): Initialize n_core_null_bytes. This is executed as part of CREATE TABLE. dict_index_build_internal_clust(): Initialize n_core_null_bytes to NO_CORE_NULL_BYTES if table->supports_instant(). row_create_index_for_mysql(): Initialize n_core_null_bytes for CREATE TEMPORARY TABLE. commit_cache_norebuild(): Call the code to rename or enlarge columns in the cache only if instant ADD COLUMN is not being used. (Instant ADD COLUMN would copy all column metadata from instant_table to old_table, including the names and lengths.) PAGE_INSTANT: A new 13-bit field for storing dict_index_t::n_core_fields. This is repurposing the 16-bit field PAGE_DIRECTION, of which only the least significant 3 bits were used. The original byte containing PAGE_DIRECTION will be accessible via the new constant PAGE_DIRECTION_B. page_get_instant(), page_set_instant(): Accessors for the PAGE_INSTANT. page_ptr_get_direction(), page_get_direction(), page_ptr_set_direction(): Accessors for PAGE_DIRECTION. page_direction_reset(): Reset PAGE_DIRECTION, PAGE_N_DIRECTION. page_direction_increment(): Increment PAGE_N_DIRECTION and set PAGE_DIRECTION. rec_get_offsets(): Use the 'leaf' parameter for non-debug purposes, and assume that heap_no is always set. Initialize all dict_index_t::n_fields for ROW_FORMAT=REDUNDANT records, even if the record contains fewer fields. rec_offs_make_valid(): Add the parameter 'leaf'. rec_copy_prefix_to_dtuple(): Assert that the tuple is only built on the core fields. Instant ADD COLUMN only applies to the clustered index, and we should never build a search key that has more than the PRIMARY KEY and possibly DB_TRX_ID,DB_ROLL_PTR. All these columns are always present. dict_index_build_data_tuple(): Remove assertions that would be duplicated in rec_copy_prefix_to_dtuple(). rec_init_offsets(): Support ROW_FORMAT=REDUNDANT records whose number of fields is between n_core_fields and n_fields. cmp_rec_rec_with_match(): Implement the comparison between two MIN_REC_FLAG records. trx_t::in_rollback: Make the field available in non-debug builds. trx_start_for_ddl_low(): Remove dangerous error-tolerance. A dictionary transaction must be flagged as such before it has generated any undo log records. This is because trx_undo_assign_undo() will mark the transaction as a dictionary transaction in the undo log header right before the very first undo log record is being written. btr_index_rec_validate(): Account for instant ADD COLUMN row_undo_ins_remove_clust_rec(): On the rollback of an insert into SYS_COLUMNS, revert instant ADD COLUMN in the cache by removing the last column from the table and the clustered index. row_search_on_row_ref(), row_undo_mod_parse_undo_rec(), row_undo_mod(), trx_undo_update_rec_get_update(): Handle the 'default row' as a special case. dtuple_t::trim(index): Omit a redundant suffix of an index tuple right before insert or update. After instant ADD COLUMN, if the last fields of a clustered index tuple match the 'default row', there is no need to store them. While trimming the entry, we must hold a page latch, so that the table cannot be emptied and the 'default row' be deleted. btr_cur_optimistic_update(), btr_cur_pessimistic_update(), row_upd_clust_rec_by_insert(), row_ins_clust_index_entry_low(): Invoke dtuple_t::trim() if needed. row_ins_clust_index_entry(): Restore dtuple_t::n_fields after calling row_ins_clust_index_entry_low(). rec_get_converted_size(), rec_get_converted_size_comp(): Allow the number of fields to be between n_core_fields and n_fields. Do not support infimum,supremum. They are never supposed to be stored in dtuple_t, because page creation nowadays uses a lower-level method for initializing them. rec_convert_dtuple_to_rec_comp(): Assign the status bits based on the number of fields. btr_cur_trim(): In an update, trim the index entry as needed. For the 'default row', handle rollback specially. For user records, omit fields that match the 'default row'. btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete(): Skip locking and adaptive hash index for the 'default row'. row_log_table_apply_convert_mrec(): Replace 'default row' values if needed. In the temporary file that is applied by row_log_table_apply(), we must identify whether the records contain the extra header for instantly added columns. For now, we will allocate an additional byte for this for ROW_T_INSERT and ROW_T_UPDATE records when the source table has been subject to instant ADD COLUMN. The ROW_T_DELETE records are fine, as they will be converted and will only contain 'core' columns (PRIMARY KEY and some system columns) that are converted from dtuple_t. rec_get_converted_size_temp(), rec_init_offsets_temp(), rec_convert_dtuple_to_temp(): Add the parameter 'status'. REC_INFO_DEFAULT_ROW = REC_INFO_MIN_REC_FLAG | REC_STATUS_COLUMNS_ADDED: An info_bits constant for distinguishing the 'default row' record. rec_comp_status_t: An enum of the status bit values. rec_leaf_format: An enum that replaces the bool parameter of rec_init_offsets_comp_ordinary().
8 years ago
MDEV-11369 Instant ADD COLUMN for InnoDB For InnoDB tables, adding, dropping and reordering columns has required a rebuild of the table and all its indexes. Since MySQL 5.6 (and MariaDB 10.0) this has been supported online (LOCK=NONE), allowing concurrent modification of the tables. This work revises the InnoDB ROW_FORMAT=REDUNDANT, ROW_FORMAT=COMPACT and ROW_FORMAT=DYNAMIC so that columns can be appended instantaneously, with only minor changes performed to the table structure. The counter innodb_instant_alter_column in INFORMATION_SCHEMA.GLOBAL_STATUS is incremented whenever a table rebuild operation is converted into an instant ADD COLUMN operation. ROW_FORMAT=COMPRESSED tables will not support instant ADD COLUMN. Some usability limitations will be addressed in subsequent work: MDEV-13134 Introduce ALTER TABLE attributes ALGORITHM=NOCOPY and ALGORITHM=INSTANT MDEV-14016 Allow instant ADD COLUMN, ADD INDEX, LOCK=NONE The format of the clustered index (PRIMARY KEY) is changed as follows: (1) The FIL_PAGE_TYPE of the root page will be FIL_PAGE_TYPE_INSTANT, and a new field PAGE_INSTANT will contain the original number of fields in the clustered index ('core' fields). If instant ADD COLUMN has not been used or the table becomes empty, or the very first instant ADD COLUMN operation is rolled back, the fields PAGE_INSTANT and FIL_PAGE_TYPE will be reset to 0 and FIL_PAGE_INDEX. (2) A special 'default row' record is inserted into the leftmost leaf, between the page infimum and the first user record. This record is distinguished by the REC_INFO_MIN_REC_FLAG, and it is otherwise in the same format as records that contain values for the instantly added columns. This 'default row' always has the same number of fields as the clustered index according to the table definition. The values of 'core' fields are to be ignored. For other fields, the 'default row' will contain the default values as they were during the ALTER TABLE statement. (If the column default values are changed later, those values will only be stored in the .frm file. The 'default row' will contain the original evaluated values, which must be the same for every row.) The 'default row' must be completely hidden from higher-level access routines. Assertions have been added to ensure that no 'default row' is ever present in the adaptive hash index or in locked records. The 'default row' is never delete-marked. (3) In clustered index leaf page records, the number of fields must reside between the number of 'core' fields (dict_index_t::n_core_fields introduced in this work) and dict_index_t::n_fields. If the number of fields is less than dict_index_t::n_fields, the missing fields are replaced with the column value of the 'default row'. Note: The number of fields in the record may shrink if some of the last instantly added columns are updated to the value that is in the 'default row'. The function btr_cur_trim() implements this 'compression' on update and rollback; dtuple::trim() implements it on insert. (4) In ROW_FORMAT=COMPACT and ROW_FORMAT=DYNAMIC records, the new status value REC_STATUS_COLUMNS_ADDED will indicate the presence of a new record header that will encode n_fields-n_core_fields-1 in 1 or 2 bytes. (In ROW_FORMAT=REDUNDANT records, the record header always explicitly encodes the number of fields.) We introduce the undo log record type TRX_UNDO_INSERT_DEFAULT for covering the insert of the 'default row' record when instant ADD COLUMN is used for the first time. Subsequent instant ADD COLUMN can use TRX_UNDO_UPD_EXIST_REC. This is joint work with Vin Chen (陈福荣) from Tencent. The design that was discussed in April 2017 would not have allowed import or export of data files, because instead of the 'default row' it would have introduced a data dictionary table. The test rpl.rpl_alter_instant is exactly as contributed in pull request #408. The test innodb.instant_alter is based on a contributed test. The redo log record format changes for ROW_FORMAT=DYNAMIC and ROW_FORMAT=COMPACT are as contributed. (With this change present, crash recovery from MariaDB 10.3.1 will fail in spectacular ways!) Also the semantics of higher-level redo log records that modify the PAGE_INSTANT field is changed. The redo log format version identifier was already changed to LOG_HEADER_FORMAT_CURRENT=103 in MariaDB 10.3.1. Everything else has been rewritten by me. Thanks to Elena Stepanova, the code has been tested extensively. When rolling back an instant ADD COLUMN operation, we must empty the PAGE_FREE list after deleting or shortening the 'default row' record, by calling either btr_page_empty() or btr_page_reorganize(). We must know the size of each entry in the PAGE_FREE list. If rollback left a freed copy of the 'default row' in the PAGE_FREE list, we would be unable to determine its size (if it is in ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC) because it would contain more fields than the rolled-back definition of the clustered index. UNIV_SQL_DEFAULT: A new special constant that designates an instantly added column that is not present in the clustered index record. len_is_stored(): Check if a length is an actual length. There are two magic length values: UNIV_SQL_DEFAULT, UNIV_SQL_NULL. dict_col_t::def_val: The 'default row' value of the column. If the column is not added instantly, def_val.len will be UNIV_SQL_DEFAULT. dict_col_t: Add the accessors is_virtual(), is_nullable(), is_instant(), instant_value(). dict_col_t::remove_instant(): Remove the 'instant ADD' status of a column. dict_col_t::name(const dict_table_t& table): Replaces dict_table_get_col_name(). dict_index_t::n_core_fields: The original number of fields. For secondary indexes and if instant ADD COLUMN has not been used, this will be equal to dict_index_t::n_fields. dict_index_t::n_core_null_bytes: Number of bytes needed to represent the null flags; usually equal to UT_BITS_IN_BYTES(n_nullable). dict_index_t::NO_CORE_NULL_BYTES: Magic value signalling that n_core_null_bytes was not initialized yet from the clustered index root page. dict_index_t: Add the accessors is_instant(), is_clust(), get_n_nullable(), instant_field_value(). dict_index_t::instant_add_field(): Adjust clustered index metadata for instant ADD COLUMN. dict_index_t::remove_instant(): Remove the 'instant ADD' status of a clustered index when the table becomes empty, or the very first instant ADD COLUMN operation is rolled back. dict_table_t: Add the accessors is_instant(), is_temporary(), supports_instant(). dict_table_t::instant_add_column(): Adjust metadata for instant ADD COLUMN. dict_table_t::rollback_instant(): Adjust metadata on the rollback of instant ADD COLUMN. prepare_inplace_alter_table_dict(): First create the ctx->new_table, and only then decide if the table really needs to be rebuilt. We must split the creation of table or index metadata from the creation of the dictionary table records and the creation of the data. In this way, we can transform a table-rebuilding operation into an instant ADD COLUMN operation. Dictionary objects will only be added to cache when table rebuilding or index creation is needed. The ctx->instant_table will never be added to cache. dict_table_t::add_to_cache(): Modified and renamed from dict_table_add_to_cache(). Do not modify the table metadata. Let the callers invoke dict_table_add_system_columns() and if needed, set can_be_evicted. dict_create_sys_tables_tuple(), dict_create_table_step(): Omit the system columns (which will now exist in the dict_table_t object already at this point). dict_create_table_step(): Expect the callers to invoke dict_table_add_system_columns(). pars_create_table(): Before creating the table creation execution graph, invoke dict_table_add_system_columns(). row_create_table_for_mysql(): Expect all callers to invoke dict_table_add_system_columns(). create_index_dict(): Replaces row_merge_create_index_graph(). innodb_update_n_cols(): Renamed from innobase_update_n_virtual(). Call my_error() if an error occurs. btr_cur_instant_init(), btr_cur_instant_init_low(), btr_cur_instant_root_init(): Load additional metadata from the clustered index and set dict_index_t::n_core_null_bytes. This is invoked when table metadata is first loaded into the data dictionary. dict_boot(): Initialize n_core_null_bytes for the four hard-coded dictionary tables. dict_create_index_step(): Initialize n_core_null_bytes. This is executed as part of CREATE TABLE. dict_index_build_internal_clust(): Initialize n_core_null_bytes to NO_CORE_NULL_BYTES if table->supports_instant(). row_create_index_for_mysql(): Initialize n_core_null_bytes for CREATE TEMPORARY TABLE. commit_cache_norebuild(): Call the code to rename or enlarge columns in the cache only if instant ADD COLUMN is not being used. (Instant ADD COLUMN would copy all column metadata from instant_table to old_table, including the names and lengths.) PAGE_INSTANT: A new 13-bit field for storing dict_index_t::n_core_fields. This is repurposing the 16-bit field PAGE_DIRECTION, of which only the least significant 3 bits were used. The original byte containing PAGE_DIRECTION will be accessible via the new constant PAGE_DIRECTION_B. page_get_instant(), page_set_instant(): Accessors for the PAGE_INSTANT. page_ptr_get_direction(), page_get_direction(), page_ptr_set_direction(): Accessors for PAGE_DIRECTION. page_direction_reset(): Reset PAGE_DIRECTION, PAGE_N_DIRECTION. page_direction_increment(): Increment PAGE_N_DIRECTION and set PAGE_DIRECTION. rec_get_offsets(): Use the 'leaf' parameter for non-debug purposes, and assume that heap_no is always set. Initialize all dict_index_t::n_fields for ROW_FORMAT=REDUNDANT records, even if the record contains fewer fields. rec_offs_make_valid(): Add the parameter 'leaf'. rec_copy_prefix_to_dtuple(): Assert that the tuple is only built on the core fields. Instant ADD COLUMN only applies to the clustered index, and we should never build a search key that has more than the PRIMARY KEY and possibly DB_TRX_ID,DB_ROLL_PTR. All these columns are always present. dict_index_build_data_tuple(): Remove assertions that would be duplicated in rec_copy_prefix_to_dtuple(). rec_init_offsets(): Support ROW_FORMAT=REDUNDANT records whose number of fields is between n_core_fields and n_fields. cmp_rec_rec_with_match(): Implement the comparison between two MIN_REC_FLAG records. trx_t::in_rollback: Make the field available in non-debug builds. trx_start_for_ddl_low(): Remove dangerous error-tolerance. A dictionary transaction must be flagged as such before it has generated any undo log records. This is because trx_undo_assign_undo() will mark the transaction as a dictionary transaction in the undo log header right before the very first undo log record is being written. btr_index_rec_validate(): Account for instant ADD COLUMN row_undo_ins_remove_clust_rec(): On the rollback of an insert into SYS_COLUMNS, revert instant ADD COLUMN in the cache by removing the last column from the table and the clustered index. row_search_on_row_ref(), row_undo_mod_parse_undo_rec(), row_undo_mod(), trx_undo_update_rec_get_update(): Handle the 'default row' as a special case. dtuple_t::trim(index): Omit a redundant suffix of an index tuple right before insert or update. After instant ADD COLUMN, if the last fields of a clustered index tuple match the 'default row', there is no need to store them. While trimming the entry, we must hold a page latch, so that the table cannot be emptied and the 'default row' be deleted. btr_cur_optimistic_update(), btr_cur_pessimistic_update(), row_upd_clust_rec_by_insert(), row_ins_clust_index_entry_low(): Invoke dtuple_t::trim() if needed. row_ins_clust_index_entry(): Restore dtuple_t::n_fields after calling row_ins_clust_index_entry_low(). rec_get_converted_size(), rec_get_converted_size_comp(): Allow the number of fields to be between n_core_fields and n_fields. Do not support infimum,supremum. They are never supposed to be stored in dtuple_t, because page creation nowadays uses a lower-level method for initializing them. rec_convert_dtuple_to_rec_comp(): Assign the status bits based on the number of fields. btr_cur_trim(): In an update, trim the index entry as needed. For the 'default row', handle rollback specially. For user records, omit fields that match the 'default row'. btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete(): Skip locking and adaptive hash index for the 'default row'. row_log_table_apply_convert_mrec(): Replace 'default row' values if needed. In the temporary file that is applied by row_log_table_apply(), we must identify whether the records contain the extra header for instantly added columns. For now, we will allocate an additional byte for this for ROW_T_INSERT and ROW_T_UPDATE records when the source table has been subject to instant ADD COLUMN. The ROW_T_DELETE records are fine, as they will be converted and will only contain 'core' columns (PRIMARY KEY and some system columns) that are converted from dtuple_t. rec_get_converted_size_temp(), rec_init_offsets_temp(), rec_convert_dtuple_to_temp(): Add the parameter 'status'. REC_INFO_DEFAULT_ROW = REC_INFO_MIN_REC_FLAG | REC_STATUS_COLUMNS_ADDED: An info_bits constant for distinguishing the 'default row' record. rec_comp_status_t: An enum of the status bit values. rec_leaf_format: An enum that replaces the bool parameter of rec_init_offsets_comp_ordinary().
8 years ago
MDEV-24971 InnoDB access freed virtual column after rollback of secondary index Problem: ======== InnoDB fails to clean the index stub if it fails to add the virtual index which contains new virtual column. But it clears the newly virtual column from index in clear_added_indexes() during inplace_alter_table. On commit, InnoDB evicts and reload the table. In case of rollback, it doesn't happen. InnoDB clears the ABORTED index while opening the table or doing the DDL. In the mean time, InnoDB can access the dropped virtual index columns while creating prebuilt or rollback of concurrent DML. Solution: ========== (1) InnoDB should maintain newly added virtual column while rollbacking the newly added virtual index. (2) InnoDB must not defer the index removal if the alter table is executed with LOCK=EXCLUSIVE. (3) For LOCK=SHARED, InnoDB should check whether the table has any other transaction lock other than alter transaction before deferring the index stub. Replaced has_new_v_col with dict_add_vcol_info in dict_index_t to indicate whether the index has any new virtual column. dict_index_t::has_new_v_col(): Returns whether the index has newly added virtual column, it doesn't say which columns are newly added virtual column ha_innobase_inplace_ctx::is_new_vcol(): Return whether the given column is added as a part of the current alter. ha_innobase_inplace_ctx::clean_new_vcol_index(): Copy the newly added virtual column to new_vcol_info in dict_index_t. Replace the column in the index fields with virtual column stored in new_vcol_info. dict_index_t::assign_new_v_col(): Store the number of virtual column added in index as a part of alter table. dict_index_t::get_n_new_vcol(): Get the number of newly added virtual column dict_index_t::assign_drop_v_col(): Allocate the memory for adding new virtual column in new_vcol_info. dict_index_t::add_drop_v_col(): Add the newly added virtual column in new_vcol_info. dict_table_t::has_lock_for_other_trx(): Whether the table has any other transaction lock than given transaction. row_merge_drop_indexes(): Add parameter alter_trx and check whether the table has any other lock than alter transaction.
5 years ago
MDEV-11369 Instant ADD COLUMN for InnoDB For InnoDB tables, adding, dropping and reordering columns has required a rebuild of the table and all its indexes. Since MySQL 5.6 (and MariaDB 10.0) this has been supported online (LOCK=NONE), allowing concurrent modification of the tables. This work revises the InnoDB ROW_FORMAT=REDUNDANT, ROW_FORMAT=COMPACT and ROW_FORMAT=DYNAMIC so that columns can be appended instantaneously, with only minor changes performed to the table structure. The counter innodb_instant_alter_column in INFORMATION_SCHEMA.GLOBAL_STATUS is incremented whenever a table rebuild operation is converted into an instant ADD COLUMN operation. ROW_FORMAT=COMPRESSED tables will not support instant ADD COLUMN. Some usability limitations will be addressed in subsequent work: MDEV-13134 Introduce ALTER TABLE attributes ALGORITHM=NOCOPY and ALGORITHM=INSTANT MDEV-14016 Allow instant ADD COLUMN, ADD INDEX, LOCK=NONE The format of the clustered index (PRIMARY KEY) is changed as follows: (1) The FIL_PAGE_TYPE of the root page will be FIL_PAGE_TYPE_INSTANT, and a new field PAGE_INSTANT will contain the original number of fields in the clustered index ('core' fields). If instant ADD COLUMN has not been used or the table becomes empty, or the very first instant ADD COLUMN operation is rolled back, the fields PAGE_INSTANT and FIL_PAGE_TYPE will be reset to 0 and FIL_PAGE_INDEX. (2) A special 'default row' record is inserted into the leftmost leaf, between the page infimum and the first user record. This record is distinguished by the REC_INFO_MIN_REC_FLAG, and it is otherwise in the same format as records that contain values for the instantly added columns. This 'default row' always has the same number of fields as the clustered index according to the table definition. The values of 'core' fields are to be ignored. For other fields, the 'default row' will contain the default values as they were during the ALTER TABLE statement. (If the column default values are changed later, those values will only be stored in the .frm file. The 'default row' will contain the original evaluated values, which must be the same for every row.) The 'default row' must be completely hidden from higher-level access routines. Assertions have been added to ensure that no 'default row' is ever present in the adaptive hash index or in locked records. The 'default row' is never delete-marked. (3) In clustered index leaf page records, the number of fields must reside between the number of 'core' fields (dict_index_t::n_core_fields introduced in this work) and dict_index_t::n_fields. If the number of fields is less than dict_index_t::n_fields, the missing fields are replaced with the column value of the 'default row'. Note: The number of fields in the record may shrink if some of the last instantly added columns are updated to the value that is in the 'default row'. The function btr_cur_trim() implements this 'compression' on update and rollback; dtuple::trim() implements it on insert. (4) In ROW_FORMAT=COMPACT and ROW_FORMAT=DYNAMIC records, the new status value REC_STATUS_COLUMNS_ADDED will indicate the presence of a new record header that will encode n_fields-n_core_fields-1 in 1 or 2 bytes. (In ROW_FORMAT=REDUNDANT records, the record header always explicitly encodes the number of fields.) We introduce the undo log record type TRX_UNDO_INSERT_DEFAULT for covering the insert of the 'default row' record when instant ADD COLUMN is used for the first time. Subsequent instant ADD COLUMN can use TRX_UNDO_UPD_EXIST_REC. This is joint work with Vin Chen (陈福荣) from Tencent. The design that was discussed in April 2017 would not have allowed import or export of data files, because instead of the 'default row' it would have introduced a data dictionary table. The test rpl.rpl_alter_instant is exactly as contributed in pull request #408. The test innodb.instant_alter is based on a contributed test. The redo log record format changes for ROW_FORMAT=DYNAMIC and ROW_FORMAT=COMPACT are as contributed. (With this change present, crash recovery from MariaDB 10.3.1 will fail in spectacular ways!) Also the semantics of higher-level redo log records that modify the PAGE_INSTANT field is changed. The redo log format version identifier was already changed to LOG_HEADER_FORMAT_CURRENT=103 in MariaDB 10.3.1. Everything else has been rewritten by me. Thanks to Elena Stepanova, the code has been tested extensively. When rolling back an instant ADD COLUMN operation, we must empty the PAGE_FREE list after deleting or shortening the 'default row' record, by calling either btr_page_empty() or btr_page_reorganize(). We must know the size of each entry in the PAGE_FREE list. If rollback left a freed copy of the 'default row' in the PAGE_FREE list, we would be unable to determine its size (if it is in ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC) because it would contain more fields than the rolled-back definition of the clustered index. UNIV_SQL_DEFAULT: A new special constant that designates an instantly added column that is not present in the clustered index record. len_is_stored(): Check if a length is an actual length. There are two magic length values: UNIV_SQL_DEFAULT, UNIV_SQL_NULL. dict_col_t::def_val: The 'default row' value of the column. If the column is not added instantly, def_val.len will be UNIV_SQL_DEFAULT. dict_col_t: Add the accessors is_virtual(), is_nullable(), is_instant(), instant_value(). dict_col_t::remove_instant(): Remove the 'instant ADD' status of a column. dict_col_t::name(const dict_table_t& table): Replaces dict_table_get_col_name(). dict_index_t::n_core_fields: The original number of fields. For secondary indexes and if instant ADD COLUMN has not been used, this will be equal to dict_index_t::n_fields. dict_index_t::n_core_null_bytes: Number of bytes needed to represent the null flags; usually equal to UT_BITS_IN_BYTES(n_nullable). dict_index_t::NO_CORE_NULL_BYTES: Magic value signalling that n_core_null_bytes was not initialized yet from the clustered index root page. dict_index_t: Add the accessors is_instant(), is_clust(), get_n_nullable(), instant_field_value(). dict_index_t::instant_add_field(): Adjust clustered index metadata for instant ADD COLUMN. dict_index_t::remove_instant(): Remove the 'instant ADD' status of a clustered index when the table becomes empty, or the very first instant ADD COLUMN operation is rolled back. dict_table_t: Add the accessors is_instant(), is_temporary(), supports_instant(). dict_table_t::instant_add_column(): Adjust metadata for instant ADD COLUMN. dict_table_t::rollback_instant(): Adjust metadata on the rollback of instant ADD COLUMN. prepare_inplace_alter_table_dict(): First create the ctx->new_table, and only then decide if the table really needs to be rebuilt. We must split the creation of table or index metadata from the creation of the dictionary table records and the creation of the data. In this way, we can transform a table-rebuilding operation into an instant ADD COLUMN operation. Dictionary objects will only be added to cache when table rebuilding or index creation is needed. The ctx->instant_table will never be added to cache. dict_table_t::add_to_cache(): Modified and renamed from dict_table_add_to_cache(). Do not modify the table metadata. Let the callers invoke dict_table_add_system_columns() and if needed, set can_be_evicted. dict_create_sys_tables_tuple(), dict_create_table_step(): Omit the system columns (which will now exist in the dict_table_t object already at this point). dict_create_table_step(): Expect the callers to invoke dict_table_add_system_columns(). pars_create_table(): Before creating the table creation execution graph, invoke dict_table_add_system_columns(). row_create_table_for_mysql(): Expect all callers to invoke dict_table_add_system_columns(). create_index_dict(): Replaces row_merge_create_index_graph(). innodb_update_n_cols(): Renamed from innobase_update_n_virtual(). Call my_error() if an error occurs. btr_cur_instant_init(), btr_cur_instant_init_low(), btr_cur_instant_root_init(): Load additional metadata from the clustered index and set dict_index_t::n_core_null_bytes. This is invoked when table metadata is first loaded into the data dictionary. dict_boot(): Initialize n_core_null_bytes for the four hard-coded dictionary tables. dict_create_index_step(): Initialize n_core_null_bytes. This is executed as part of CREATE TABLE. dict_index_build_internal_clust(): Initialize n_core_null_bytes to NO_CORE_NULL_BYTES if table->supports_instant(). row_create_index_for_mysql(): Initialize n_core_null_bytes for CREATE TEMPORARY TABLE. commit_cache_norebuild(): Call the code to rename or enlarge columns in the cache only if instant ADD COLUMN is not being used. (Instant ADD COLUMN would copy all column metadata from instant_table to old_table, including the names and lengths.) PAGE_INSTANT: A new 13-bit field for storing dict_index_t::n_core_fields. This is repurposing the 16-bit field PAGE_DIRECTION, of which only the least significant 3 bits were used. The original byte containing PAGE_DIRECTION will be accessible via the new constant PAGE_DIRECTION_B. page_get_instant(), page_set_instant(): Accessors for the PAGE_INSTANT. page_ptr_get_direction(), page_get_direction(), page_ptr_set_direction(): Accessors for PAGE_DIRECTION. page_direction_reset(): Reset PAGE_DIRECTION, PAGE_N_DIRECTION. page_direction_increment(): Increment PAGE_N_DIRECTION and set PAGE_DIRECTION. rec_get_offsets(): Use the 'leaf' parameter for non-debug purposes, and assume that heap_no is always set. Initialize all dict_index_t::n_fields for ROW_FORMAT=REDUNDANT records, even if the record contains fewer fields. rec_offs_make_valid(): Add the parameter 'leaf'. rec_copy_prefix_to_dtuple(): Assert that the tuple is only built on the core fields. Instant ADD COLUMN only applies to the clustered index, and we should never build a search key that has more than the PRIMARY KEY and possibly DB_TRX_ID,DB_ROLL_PTR. All these columns are always present. dict_index_build_data_tuple(): Remove assertions that would be duplicated in rec_copy_prefix_to_dtuple(). rec_init_offsets(): Support ROW_FORMAT=REDUNDANT records whose number of fields is between n_core_fields and n_fields. cmp_rec_rec_with_match(): Implement the comparison between two MIN_REC_FLAG records. trx_t::in_rollback: Make the field available in non-debug builds. trx_start_for_ddl_low(): Remove dangerous error-tolerance. A dictionary transaction must be flagged as such before it has generated any undo log records. This is because trx_undo_assign_undo() will mark the transaction as a dictionary transaction in the undo log header right before the very first undo log record is being written. btr_index_rec_validate(): Account for instant ADD COLUMN row_undo_ins_remove_clust_rec(): On the rollback of an insert into SYS_COLUMNS, revert instant ADD COLUMN in the cache by removing the last column from the table and the clustered index. row_search_on_row_ref(), row_undo_mod_parse_undo_rec(), row_undo_mod(), trx_undo_update_rec_get_update(): Handle the 'default row' as a special case. dtuple_t::trim(index): Omit a redundant suffix of an index tuple right before insert or update. After instant ADD COLUMN, if the last fields of a clustered index tuple match the 'default row', there is no need to store them. While trimming the entry, we must hold a page latch, so that the table cannot be emptied and the 'default row' be deleted. btr_cur_optimistic_update(), btr_cur_pessimistic_update(), row_upd_clust_rec_by_insert(), row_ins_clust_index_entry_low(): Invoke dtuple_t::trim() if needed. row_ins_clust_index_entry(): Restore dtuple_t::n_fields after calling row_ins_clust_index_entry_low(). rec_get_converted_size(), rec_get_converted_size_comp(): Allow the number of fields to be between n_core_fields and n_fields. Do not support infimum,supremum. They are never supposed to be stored in dtuple_t, because page creation nowadays uses a lower-level method for initializing them. rec_convert_dtuple_to_rec_comp(): Assign the status bits based on the number of fields. btr_cur_trim(): In an update, trim the index entry as needed. For the 'default row', handle rollback specially. For user records, omit fields that match the 'default row'. btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete(): Skip locking and adaptive hash index for the 'default row'. row_log_table_apply_convert_mrec(): Replace 'default row' values if needed. In the temporary file that is applied by row_log_table_apply(), we must identify whether the records contain the extra header for instantly added columns. For now, we will allocate an additional byte for this for ROW_T_INSERT and ROW_T_UPDATE records when the source table has been subject to instant ADD COLUMN. The ROW_T_DELETE records are fine, as they will be converted and will only contain 'core' columns (PRIMARY KEY and some system columns) that are converted from dtuple_t. rec_get_converted_size_temp(), rec_init_offsets_temp(), rec_convert_dtuple_to_temp(): Add the parameter 'status'. REC_INFO_DEFAULT_ROW = REC_INFO_MIN_REC_FLAG | REC_STATUS_COLUMNS_ADDED: An info_bits constant for distinguishing the 'default row' record. rec_comp_status_t: An enum of the status bit values. rec_leaf_format: An enum that replaces the bool parameter of rec_init_offsets_comp_ordinary().
8 years ago
MDEV-11369 Instant ADD COLUMN for InnoDB For InnoDB tables, adding, dropping and reordering columns has required a rebuild of the table and all its indexes. Since MySQL 5.6 (and MariaDB 10.0) this has been supported online (LOCK=NONE), allowing concurrent modification of the tables. This work revises the InnoDB ROW_FORMAT=REDUNDANT, ROW_FORMAT=COMPACT and ROW_FORMAT=DYNAMIC so that columns can be appended instantaneously, with only minor changes performed to the table structure. The counter innodb_instant_alter_column in INFORMATION_SCHEMA.GLOBAL_STATUS is incremented whenever a table rebuild operation is converted into an instant ADD COLUMN operation. ROW_FORMAT=COMPRESSED tables will not support instant ADD COLUMN. Some usability limitations will be addressed in subsequent work: MDEV-13134 Introduce ALTER TABLE attributes ALGORITHM=NOCOPY and ALGORITHM=INSTANT MDEV-14016 Allow instant ADD COLUMN, ADD INDEX, LOCK=NONE The format of the clustered index (PRIMARY KEY) is changed as follows: (1) The FIL_PAGE_TYPE of the root page will be FIL_PAGE_TYPE_INSTANT, and a new field PAGE_INSTANT will contain the original number of fields in the clustered index ('core' fields). If instant ADD COLUMN has not been used or the table becomes empty, or the very first instant ADD COLUMN operation is rolled back, the fields PAGE_INSTANT and FIL_PAGE_TYPE will be reset to 0 and FIL_PAGE_INDEX. (2) A special 'default row' record is inserted into the leftmost leaf, between the page infimum and the first user record. This record is distinguished by the REC_INFO_MIN_REC_FLAG, and it is otherwise in the same format as records that contain values for the instantly added columns. This 'default row' always has the same number of fields as the clustered index according to the table definition. The values of 'core' fields are to be ignored. For other fields, the 'default row' will contain the default values as they were during the ALTER TABLE statement. (If the column default values are changed later, those values will only be stored in the .frm file. The 'default row' will contain the original evaluated values, which must be the same for every row.) The 'default row' must be completely hidden from higher-level access routines. Assertions have been added to ensure that no 'default row' is ever present in the adaptive hash index or in locked records. The 'default row' is never delete-marked. (3) In clustered index leaf page records, the number of fields must reside between the number of 'core' fields (dict_index_t::n_core_fields introduced in this work) and dict_index_t::n_fields. If the number of fields is less than dict_index_t::n_fields, the missing fields are replaced with the column value of the 'default row'. Note: The number of fields in the record may shrink if some of the last instantly added columns are updated to the value that is in the 'default row'. The function btr_cur_trim() implements this 'compression' on update and rollback; dtuple::trim() implements it on insert. (4) In ROW_FORMAT=COMPACT and ROW_FORMAT=DYNAMIC records, the new status value REC_STATUS_COLUMNS_ADDED will indicate the presence of a new record header that will encode n_fields-n_core_fields-1 in 1 or 2 bytes. (In ROW_FORMAT=REDUNDANT records, the record header always explicitly encodes the number of fields.) We introduce the undo log record type TRX_UNDO_INSERT_DEFAULT for covering the insert of the 'default row' record when instant ADD COLUMN is used for the first time. Subsequent instant ADD COLUMN can use TRX_UNDO_UPD_EXIST_REC. This is joint work with Vin Chen (陈福荣) from Tencent. The design that was discussed in April 2017 would not have allowed import or export of data files, because instead of the 'default row' it would have introduced a data dictionary table. The test rpl.rpl_alter_instant is exactly as contributed in pull request #408. The test innodb.instant_alter is based on a contributed test. The redo log record format changes for ROW_FORMAT=DYNAMIC and ROW_FORMAT=COMPACT are as contributed. (With this change present, crash recovery from MariaDB 10.3.1 will fail in spectacular ways!) Also the semantics of higher-level redo log records that modify the PAGE_INSTANT field is changed. The redo log format version identifier was already changed to LOG_HEADER_FORMAT_CURRENT=103 in MariaDB 10.3.1. Everything else has been rewritten by me. Thanks to Elena Stepanova, the code has been tested extensively. When rolling back an instant ADD COLUMN operation, we must empty the PAGE_FREE list after deleting or shortening the 'default row' record, by calling either btr_page_empty() or btr_page_reorganize(). We must know the size of each entry in the PAGE_FREE list. If rollback left a freed copy of the 'default row' in the PAGE_FREE list, we would be unable to determine its size (if it is in ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC) because it would contain more fields than the rolled-back definition of the clustered index. UNIV_SQL_DEFAULT: A new special constant that designates an instantly added column that is not present in the clustered index record. len_is_stored(): Check if a length is an actual length. There are two magic length values: UNIV_SQL_DEFAULT, UNIV_SQL_NULL. dict_col_t::def_val: The 'default row' value of the column. If the column is not added instantly, def_val.len will be UNIV_SQL_DEFAULT. dict_col_t: Add the accessors is_virtual(), is_nullable(), is_instant(), instant_value(). dict_col_t::remove_instant(): Remove the 'instant ADD' status of a column. dict_col_t::name(const dict_table_t& table): Replaces dict_table_get_col_name(). dict_index_t::n_core_fields: The original number of fields. For secondary indexes and if instant ADD COLUMN has not been used, this will be equal to dict_index_t::n_fields. dict_index_t::n_core_null_bytes: Number of bytes needed to represent the null flags; usually equal to UT_BITS_IN_BYTES(n_nullable). dict_index_t::NO_CORE_NULL_BYTES: Magic value signalling that n_core_null_bytes was not initialized yet from the clustered index root page. dict_index_t: Add the accessors is_instant(), is_clust(), get_n_nullable(), instant_field_value(). dict_index_t::instant_add_field(): Adjust clustered index metadata for instant ADD COLUMN. dict_index_t::remove_instant(): Remove the 'instant ADD' status of a clustered index when the table becomes empty, or the very first instant ADD COLUMN operation is rolled back. dict_table_t: Add the accessors is_instant(), is_temporary(), supports_instant(). dict_table_t::instant_add_column(): Adjust metadata for instant ADD COLUMN. dict_table_t::rollback_instant(): Adjust metadata on the rollback of instant ADD COLUMN. prepare_inplace_alter_table_dict(): First create the ctx->new_table, and only then decide if the table really needs to be rebuilt. We must split the creation of table or index metadata from the creation of the dictionary table records and the creation of the data. In this way, we can transform a table-rebuilding operation into an instant ADD COLUMN operation. Dictionary objects will only be added to cache when table rebuilding or index creation is needed. The ctx->instant_table will never be added to cache. dict_table_t::add_to_cache(): Modified and renamed from dict_table_add_to_cache(). Do not modify the table metadata. Let the callers invoke dict_table_add_system_columns() and if needed, set can_be_evicted. dict_create_sys_tables_tuple(), dict_create_table_step(): Omit the system columns (which will now exist in the dict_table_t object already at this point). dict_create_table_step(): Expect the callers to invoke dict_table_add_system_columns(). pars_create_table(): Before creating the table creation execution graph, invoke dict_table_add_system_columns(). row_create_table_for_mysql(): Expect all callers to invoke dict_table_add_system_columns(). create_index_dict(): Replaces row_merge_create_index_graph(). innodb_update_n_cols(): Renamed from innobase_update_n_virtual(). Call my_error() if an error occurs. btr_cur_instant_init(), btr_cur_instant_init_low(), btr_cur_instant_root_init(): Load additional metadata from the clustered index and set dict_index_t::n_core_null_bytes. This is invoked when table metadata is first loaded into the data dictionary. dict_boot(): Initialize n_core_null_bytes for the four hard-coded dictionary tables. dict_create_index_step(): Initialize n_core_null_bytes. This is executed as part of CREATE TABLE. dict_index_build_internal_clust(): Initialize n_core_null_bytes to NO_CORE_NULL_BYTES if table->supports_instant(). row_create_index_for_mysql(): Initialize n_core_null_bytes for CREATE TEMPORARY TABLE. commit_cache_norebuild(): Call the code to rename or enlarge columns in the cache only if instant ADD COLUMN is not being used. (Instant ADD COLUMN would copy all column metadata from instant_table to old_table, including the names and lengths.) PAGE_INSTANT: A new 13-bit field for storing dict_index_t::n_core_fields. This is repurposing the 16-bit field PAGE_DIRECTION, of which only the least significant 3 bits were used. The original byte containing PAGE_DIRECTION will be accessible via the new constant PAGE_DIRECTION_B. page_get_instant(), page_set_instant(): Accessors for the PAGE_INSTANT. page_ptr_get_direction(), page_get_direction(), page_ptr_set_direction(): Accessors for PAGE_DIRECTION. page_direction_reset(): Reset PAGE_DIRECTION, PAGE_N_DIRECTION. page_direction_increment(): Increment PAGE_N_DIRECTION and set PAGE_DIRECTION. rec_get_offsets(): Use the 'leaf' parameter for non-debug purposes, and assume that heap_no is always set. Initialize all dict_index_t::n_fields for ROW_FORMAT=REDUNDANT records, even if the record contains fewer fields. rec_offs_make_valid(): Add the parameter 'leaf'. rec_copy_prefix_to_dtuple(): Assert that the tuple is only built on the core fields. Instant ADD COLUMN only applies to the clustered index, and we should never build a search key that has more than the PRIMARY KEY and possibly DB_TRX_ID,DB_ROLL_PTR. All these columns are always present. dict_index_build_data_tuple(): Remove assertions that would be duplicated in rec_copy_prefix_to_dtuple(). rec_init_offsets(): Support ROW_FORMAT=REDUNDANT records whose number of fields is between n_core_fields and n_fields. cmp_rec_rec_with_match(): Implement the comparison between two MIN_REC_FLAG records. trx_t::in_rollback: Make the field available in non-debug builds. trx_start_for_ddl_low(): Remove dangerous error-tolerance. A dictionary transaction must be flagged as such before it has generated any undo log records. This is because trx_undo_assign_undo() will mark the transaction as a dictionary transaction in the undo log header right before the very first undo log record is being written. btr_index_rec_validate(): Account for instant ADD COLUMN row_undo_ins_remove_clust_rec(): On the rollback of an insert into SYS_COLUMNS, revert instant ADD COLUMN in the cache by removing the last column from the table and the clustered index. row_search_on_row_ref(), row_undo_mod_parse_undo_rec(), row_undo_mod(), trx_undo_update_rec_get_update(): Handle the 'default row' as a special case. dtuple_t::trim(index): Omit a redundant suffix of an index tuple right before insert or update. After instant ADD COLUMN, if the last fields of a clustered index tuple match the 'default row', there is no need to store them. While trimming the entry, we must hold a page latch, so that the table cannot be emptied and the 'default row' be deleted. btr_cur_optimistic_update(), btr_cur_pessimistic_update(), row_upd_clust_rec_by_insert(), row_ins_clust_index_entry_low(): Invoke dtuple_t::trim() if needed. row_ins_clust_index_entry(): Restore dtuple_t::n_fields after calling row_ins_clust_index_entry_low(). rec_get_converted_size(), rec_get_converted_size_comp(): Allow the number of fields to be between n_core_fields and n_fields. Do not support infimum,supremum. They are never supposed to be stored in dtuple_t, because page creation nowadays uses a lower-level method for initializing them. rec_convert_dtuple_to_rec_comp(): Assign the status bits based on the number of fields. btr_cur_trim(): In an update, trim the index entry as needed. For the 'default row', handle rollback specially. For user records, omit fields that match the 'default row'. btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete(): Skip locking and adaptive hash index for the 'default row'. row_log_table_apply_convert_mrec(): Replace 'default row' values if needed. In the temporary file that is applied by row_log_table_apply(), we must identify whether the records contain the extra header for instantly added columns. For now, we will allocate an additional byte for this for ROW_T_INSERT and ROW_T_UPDATE records when the source table has been subject to instant ADD COLUMN. The ROW_T_DELETE records are fine, as they will be converted and will only contain 'core' columns (PRIMARY KEY and some system columns) that are converted from dtuple_t. rec_get_converted_size_temp(), rec_init_offsets_temp(), rec_convert_dtuple_to_temp(): Add the parameter 'status'. REC_INFO_DEFAULT_ROW = REC_INFO_MIN_REC_FLAG | REC_STATUS_COLUMNS_ADDED: An info_bits constant for distinguishing the 'default row' record. rec_comp_status_t: An enum of the status bit values. rec_leaf_format: An enum that replaces the bool parameter of rec_init_offsets_comp_ordinary().
8 years ago
MDEV-15662 Instant DROP COLUMN or changing the order of columns Allow ADD COLUMN anywhere in a table, not only adding as the last column. Allow instant DROP COLUMN and instant changing the order of columns. The added columns will always be added last in clustered index records. In new records, instantly dropped columns will be stored as NULL or empty when possible. Information about dropped and reordered columns will be written in a metadata BLOB (mblob), which is stored before the first 'user' field in the hidden metadata record at the start of the clustered index. The presence of mblob is indicated by setting the delete-mark flag in the metadata record. The metadata BLOB stores the number of clustered index fields, followed by an array of column information for each field. For dropped columns, we store the NOT NULL flag, the fixed length, and for variable-length columns, whether the maximum length exceeded 255 bytes. For non-dropped columns, we store the column position. Unlike with MDEV-11369, when a table becomes empty, it cannot be converted back to the canonical format. The reason for this is that other threads may hold cached objects such as row_prebuilt_t::ins_node that could refer to dropped or reordered index fields. For instant DROP COLUMN and ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC, we must store the n_core_null_bytes in the root page, so that the chain of node pointer records can be followed in order to reach the leftmost leaf page where the metadata record is located. If the mblob is present, we will zero-initialize the strings "infimum" and "supremum" in the root page, and use the last byte of "supremum" for storing the number of null bytes (which are allocated but useless on node pointer pages). This is necessary for btr_cur_instant_init_metadata() to be able to navigate to the mblob. If the PRIMARY KEY contains any variable-length column and some nullable columns were instantly dropped, the dict_index_t::n_nullable in the data dictionary could be smaller than it actually is in the non-leaf pages. Because of this, the non-leaf pages could use more bytes for the null flags than the data dictionary expects, and we could be reading the lengths of the variable-length columns from the wrong offset, and thus reading the child page number from wrong place. This is the result of two design mistakes that involve unnecessary storage of data: First, it is nonsense to store any data fields for the leftmost node pointer records, because the comparisons would be resolved by the MIN_REC_FLAG alone. Second, there cannot be any null fields in the clustered index node pointer fields, but we nevertheless reserve space for all the null flags. Limitations (future work): MDEV-17459 Allow instant ALTER TABLE even if FULLTEXT INDEX exists MDEV-17468 Avoid table rebuild on operations on generated columns MDEV-17494 Refuse ALGORITHM=INSTANT when the row size is too large btr_page_reorganize_low(): Preserve any metadata in the root page. Call lock_move_reorganize_page() only after restoring the "infimum" and "supremum" records, to avoid a memcmp() assertion failure. dict_col_t::DROPPED: Magic value for dict_col_t::ind. dict_col_t::clear_instant(): Renamed from dict_col_t::remove_instant(). Do not assert that the column was instantly added, because we sometimes call this unconditionally for all columns. Convert an instantly added column to a "core column". The old name remove_instant() could be mistaken to refer to "instant DROP COLUMN". dict_col_t::is_added(): Rename from dict_col_t::is_instant(). dtype_t::metadata_blob_init(): Initialize the mblob data type. dtuple_t::is_metadata(), dtuple_t::is_alter_metadata(), upd_t::is_metadata(), upd_t::is_alter_metadata(): Check if info_bits refer to a metadata record. dict_table_t::instant: Metadata about dropped or reordered columns. dict_table_t::prepare_instant(): Prepare ha_innobase_inplace_ctx::instant_table for instant ALTER TABLE. innobase_instant_try() will pass this to dict_table_t::instant_column(). On rollback, dict_table_t::rollback_instant() will be called. dict_table_t::instant_column(): Renamed from instant_add_column(). Add the parameter col_map so that columns can be reordered. Copy and adjust v_cols[] as well. dict_table_t::find(): Find an old column based on a new column number. dict_table_t::serialise_columns(), dict_table_t::deserialise_columns(): Convert the mblob. dict_index_t::instant_metadata(): Create the metadata record for instant ALTER TABLE. Invoke dict_table_t::serialise_columns(). dict_index_t::reconstruct_fields(): Invoked by dict_table_t::deserialise_columns(). dict_index_t::clear_instant_alter(): Move the fields for the dropped columns to the end, and sort the surviving index fields in ascending order of column position. ha_innobase::check_if_supported_inplace_alter(): Do not allow adding a FTS_DOC_ID column if a hidden FTS_DOC_ID column exists due to FULLTEXT INDEX. (This always required ALGORITHM=COPY.) instant_alter_column_possible(): Add a parameter for InnoDB table, to check for additional conditions, such as the maximum number of index fields. ha_innobase_inplace_ctx::first_alter_pos: The first column whose position is affected by instant ADD, DROP, or changing the order of columns. innobase_build_col_map(): Skip added virtual columns. prepare_inplace_add_virtual(): Correctly compute num_to_add_vcol. Remove some unnecessary code. Note that the call to innodb_base_col_setup() should be executed later. commit_try_norebuild(): If ctx->is_instant(), let the virtual columns be added or dropped by innobase_instant_try(). innobase_instant_try(): Fill in a zero default value for the hidden column FTS_DOC_ID (to reduce the work needed in MDEV-17459). If any columns were dropped or reordered (or added not last), delete any SYS_COLUMNS records for the following columns, and insert SYS_COLUMNS records for all subsequent stored columns as well as for all virtual columns. If any virtual column is dropped, rewrite all virtual column metadata. Use a shortcut only for adding virtual columns. This is because innobase_drop_virtual_try() assumes that the dropped virtual columns still exist in ctx->old_table. innodb_update_cols(): Renamed from innodb_update_n_cols(). innobase_add_one_virtual(), innobase_insert_sys_virtual(): Change the return type to bool, and invoke my_error() when detecting an error. innodb_insert_sys_columns(): Insert a record into SYS_COLUMNS. Refactored from innobase_add_one_virtual() and innobase_instant_add_col(). innobase_instant_add_col(): Replace the parameter dfield with type. innobase_instant_drop_cols(): Drop matching columns from SYS_COLUMNS and all columns from SYS_VIRTUAL. innobase_add_virtual_try(), innobase_drop_virtual_try(): Let the caller invoke innodb_update_cols(). innobase_rename_column_try(): Skip dropped columns. commit_cache_norebuild(): Update table->fts->doc_col. dict_mem_table_col_rename_low(): Skip dropped columns. trx_undo_rec_get_partial_row(): Skip dropped columns. trx_undo_update_rec_get_update(): Handle the metadata BLOB correctly. trx_undo_page_report_modify(): Avoid out-of-bounds access to record fields. Log metadata records consistently. Apparently, the first fields of a clustered index may be updated in an update_undo vector when the index is ID_IND of SYS_FOREIGN, as part of renaming the table during ALTER TABLE. Normally, updates of the PRIMARY KEY should be logged as delete-mark and an insert. row_undo_mod_parse_undo_rec(), row_purge_parse_undo_rec(): Use trx_undo_metadata. row_undo_mod_clust_low(): On metadata rollback, roll back the root page too. row_undo_mod_clust(): Relax an assertion. The delete-mark flag was repurposed for ALTER TABLE metadata records. row_rec_to_index_entry_impl(): Add the template parameter mblob and the optional parameter info_bits for specifying the desired new info bits. For the metadata tuple, allow conversion between the original format (ADD COLUMN only) and the generic format (with hidden BLOB). Add the optional parameter "pad" to determine whether the tuple should be padded to the index fields (on ALTER TABLE it should), or whether it should remain at its original size (on rollback). row_build_index_entry_low(): Clean up the code, removing redundant variables and conditions. For instantly dropped columns, generate a dummy value that is NULL, the empty string, or a fixed length of NUL bytes, depending on the type of the dropped column. row_upd_clust_rec_by_insert_inherit_func(): On the update of PRIMARY KEY of a record that contained a dropped column whose value was stored externally, we will be inserting a dummy NULL or empty string value to the field of the dropped column. The externally stored column would eventually be dropped when purge removes the delete-marked record for the old PRIMARY KEY value. btr_index_rec_validate(): Recognize the metadata record. btr_discard_only_page_on_level(): Preserve the generic instant ALTER TABLE metadata. btr_set_instant(): Replaces page_set_instant(). This sets a clustered index root page to the appropriate format, or upgrades from the MDEV-11369 instant ADD COLUMN to generic ALTER TABLE format. btr_cur_instant_init_low(): Read and validate the metadata BLOB page before reconstructing the dictionary information based on it. btr_cur_instant_init_metadata(): Do not read any lengths from the metadata record header before reading the BLOB. At this point, we would not actually know how many nullable fields the metadata record contains. btr_cur_instant_root_init(): Initialize n_core_null_bytes in one of two possible ways. btr_cur_trim(): Handle the mblob record. row_metadata_to_tuple(): Convert a metadata record to a data tuple, based on the new info_bits of the metadata record. btr_cur_pessimistic_update(): Invoke row_metadata_to_tuple() if needed. Invoke dtuple_convert_big_rec() for metadata records if the record is too large, or if the mblob is not yet marked as externally stored. btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete(): When the last user record is deleted, do not delete the generic instant ALTER TABLE metadata record. Only delete MDEV-11369 instant ADD COLUMN metadata records. btr_cur_optimistic_insert(): Avoid unnecessary computation of rec_size. btr_pcur_store_position(): Allow a logically empty page to contain a metadata record for generic ALTER TABLE. REC_INFO_DEFAULT_ROW_ADD: Renamed from REC_INFO_DEFAULT_ROW. This is for the old instant ADD COLUMN (MDEV-11369) only. REC_INFO_DEFAULT_ROW_ALTER: The more generic metadata record, with additional information for dropped or reordered columns. rec_info_bits_valid(): Remove. The only case when this would fail is when the record is the generic ALTER TABLE metadata record. rec_is_alter_metadata(): Check if a record is the metadata record for instant ALTER TABLE (other than ADD COLUMN). NOTE: This function must not be invoked on node pointer records, because the delete-mark flag in those records may be set (it is garbage), and then a debug assertion could fail because index->is_instant() does not necessarily hold. rec_is_add_metadata(): Check if a record is MDEV-11369 ADD COLUMN metadata record (not more generic instant ALTER TABLE). rec_get_converted_size_comp_prefix_low(): Assume that the metadata field will be stored externally. In dtuple_convert_big_rec() during the rec_get_converted_size() call, it would not be there yet. rec_get_converted_size_comp(): Replace status,fields,n_fields with tuple. rec_init_offsets_comp_ordinary(), rec_get_converted_size_comp_prefix_low(), rec_convert_dtuple_to_rec_comp(): Add template<bool mblob = false>. With mblob=true, process a record with a metadata BLOB. rec_copy_prefix_to_buf(): Assert that no fields beyond the key and system columns are being copied. Exclude the metadata BLOB field. rec_convert_dtuple_to_metadata_comp(): Convert an alter metadata tuple into a record. row_upd_index_replace_metadata(): Apply an update vector to an alter_metadata tuple. row_log_allocate(): Replace dict_index_t::is_instant() with a more appropriate condition that ignores dict_table_t::instant. Only a table on which the MDEV-11369 ADD COLUMN was performed can "lose its instantness" when it becomes empty. After instant DROP COLUMN or reordering columns, we cannot simply convert the table to the canonical format, because the data dictionary cache and all possibly existing references to it from other client connection threads would have to be adjusted. row_quiesce_write_index_fields(): Do not crash when the table contains an instantly dropped column. Thanks to Thirunarayanan Balathandayuthapani for discussing the design and implementing an initial prototype of this. Thanks to Matthias Leich for testing.
7 years ago
MDEV-22456 Dropping the adaptive hash index may cause DDL to lock up InnoDB If the InnoDB buffer pool contains many pages for a table or index that is being dropped or rebuilt, and if many of such pages are pointed to by the adaptive hash index, dropping the adaptive hash index may consume a lot of time. The time-consuming operation of dropping the adaptive hash index entries is being executed while the InnoDB data dictionary cache dict_sys is exclusively locked. It is not actually necessary to drop all adaptive hash index entries at the time a table or index is being dropped or rebuilt. We can let the LRU replacement policy of the buffer pool take care of this gradually. For this to work, we must detach the dict_table_t and dict_index_t objects from the main dict_sys cache, and once the last adaptive hash index entry for the detached table is removed (when the garbage page is evicted from the buffer pool) we can free the dict_table_t and dict_index_t object. Related to this, in MDEV-16283, we made ALTER TABLE...DISCARD TABLESPACE skip both the buffer pool eviction and the drop of the adaptive hash index. We shifted the burden to ALTER TABLE...IMPORT TABLESPACE or DROP TABLE. We can remove the eviction from DROP TABLE. We must retain the eviction in the ALTER TABLE...IMPORT TABLESPACE code path, so that in case the discarded table is being re-imported with the same tablespace identifier, the fresh data from the imported tablespace will replace any stale pages in the buffer pool. rpl.rpl_failed_drop_tbl_binlog: Remove the test. DROP TABLE can no longer be interrupted inside InnoDB. fseg_free_page(), fseg_free_step(), fseg_free_step_not_header(), fseg_free_page_low(), fseg_free_extent(): Remove the parameter that specifies whether the adaptive hash index should be dropped. btr_search_lazy_free(): Lazily free an index when the last reference to it is dropped from the adaptive hash index. buf_pool_clear_hash_index(): Declare static, and move to the same compilation unit with the bulk of the adaptive hash index code. dict_index_t::clone(), dict_index_t::clone_if_needed(): Clone an index that is being rebuilt while adaptive hash index entries exist. The original index will be inserted into dict_table_t::freed_indexes and dict_index_t::set_freed() will be called. dict_index_t::set_freed(), dict_index_t::freed(): Note that or check whether the index has been freed. We will use the impossible page number 1 to denote this condition. dict_index_t::n_ahi_pages(): Replaces btr_search_info_get_ref_count(). dict_index_t::detach_columns(): Move the assignment n_fields=0 to ha_innobase_inplace_ctx::clear_added_indexes(). We must have access to the columns when freeing the adaptive hash index. Note: dict_table_t::v_cols[] will remain valid. If virtual columns are dropped or added, the table definition will be reloaded in ha_innobase::commit_inplace_alter_table(). buf_page_mtr_lock(): Drop a stale adaptive hash index if needed. We will also reduce the number of btr_get_search_latch() calls and enclose some more code inside #ifdef BTR_CUR_HASH_ADAPT in order to benefit cmake -DWITH_INNODB_AHI=OFF.
5 years ago
MDEV-11369 Instant ADD COLUMN for InnoDB For InnoDB tables, adding, dropping and reordering columns has required a rebuild of the table and all its indexes. Since MySQL 5.6 (and MariaDB 10.0) this has been supported online (LOCK=NONE), allowing concurrent modification of the tables. This work revises the InnoDB ROW_FORMAT=REDUNDANT, ROW_FORMAT=COMPACT and ROW_FORMAT=DYNAMIC so that columns can be appended instantaneously, with only minor changes performed to the table structure. The counter innodb_instant_alter_column in INFORMATION_SCHEMA.GLOBAL_STATUS is incremented whenever a table rebuild operation is converted into an instant ADD COLUMN operation. ROW_FORMAT=COMPRESSED tables will not support instant ADD COLUMN. Some usability limitations will be addressed in subsequent work: MDEV-13134 Introduce ALTER TABLE attributes ALGORITHM=NOCOPY and ALGORITHM=INSTANT MDEV-14016 Allow instant ADD COLUMN, ADD INDEX, LOCK=NONE The format of the clustered index (PRIMARY KEY) is changed as follows: (1) The FIL_PAGE_TYPE of the root page will be FIL_PAGE_TYPE_INSTANT, and a new field PAGE_INSTANT will contain the original number of fields in the clustered index ('core' fields). If instant ADD COLUMN has not been used or the table becomes empty, or the very first instant ADD COLUMN operation is rolled back, the fields PAGE_INSTANT and FIL_PAGE_TYPE will be reset to 0 and FIL_PAGE_INDEX. (2) A special 'default row' record is inserted into the leftmost leaf, between the page infimum and the first user record. This record is distinguished by the REC_INFO_MIN_REC_FLAG, and it is otherwise in the same format as records that contain values for the instantly added columns. This 'default row' always has the same number of fields as the clustered index according to the table definition. The values of 'core' fields are to be ignored. For other fields, the 'default row' will contain the default values as they were during the ALTER TABLE statement. (If the column default values are changed later, those values will only be stored in the .frm file. The 'default row' will contain the original evaluated values, which must be the same for every row.) The 'default row' must be completely hidden from higher-level access routines. Assertions have been added to ensure that no 'default row' is ever present in the adaptive hash index or in locked records. The 'default row' is never delete-marked. (3) In clustered index leaf page records, the number of fields must reside between the number of 'core' fields (dict_index_t::n_core_fields introduced in this work) and dict_index_t::n_fields. If the number of fields is less than dict_index_t::n_fields, the missing fields are replaced with the column value of the 'default row'. Note: The number of fields in the record may shrink if some of the last instantly added columns are updated to the value that is in the 'default row'. The function btr_cur_trim() implements this 'compression' on update and rollback; dtuple::trim() implements it on insert. (4) In ROW_FORMAT=COMPACT and ROW_FORMAT=DYNAMIC records, the new status value REC_STATUS_COLUMNS_ADDED will indicate the presence of a new record header that will encode n_fields-n_core_fields-1 in 1 or 2 bytes. (In ROW_FORMAT=REDUNDANT records, the record header always explicitly encodes the number of fields.) We introduce the undo log record type TRX_UNDO_INSERT_DEFAULT for covering the insert of the 'default row' record when instant ADD COLUMN is used for the first time. Subsequent instant ADD COLUMN can use TRX_UNDO_UPD_EXIST_REC. This is joint work with Vin Chen (陈福荣) from Tencent. The design that was discussed in April 2017 would not have allowed import or export of data files, because instead of the 'default row' it would have introduced a data dictionary table. The test rpl.rpl_alter_instant is exactly as contributed in pull request #408. The test innodb.instant_alter is based on a contributed test. The redo log record format changes for ROW_FORMAT=DYNAMIC and ROW_FORMAT=COMPACT are as contributed. (With this change present, crash recovery from MariaDB 10.3.1 will fail in spectacular ways!) Also the semantics of higher-level redo log records that modify the PAGE_INSTANT field is changed. The redo log format version identifier was already changed to LOG_HEADER_FORMAT_CURRENT=103 in MariaDB 10.3.1. Everything else has been rewritten by me. Thanks to Elena Stepanova, the code has been tested extensively. When rolling back an instant ADD COLUMN operation, we must empty the PAGE_FREE list after deleting or shortening the 'default row' record, by calling either btr_page_empty() or btr_page_reorganize(). We must know the size of each entry in the PAGE_FREE list. If rollback left a freed copy of the 'default row' in the PAGE_FREE list, we would be unable to determine its size (if it is in ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC) because it would contain more fields than the rolled-back definition of the clustered index. UNIV_SQL_DEFAULT: A new special constant that designates an instantly added column that is not present in the clustered index record. len_is_stored(): Check if a length is an actual length. There are two magic length values: UNIV_SQL_DEFAULT, UNIV_SQL_NULL. dict_col_t::def_val: The 'default row' value of the column. If the column is not added instantly, def_val.len will be UNIV_SQL_DEFAULT. dict_col_t: Add the accessors is_virtual(), is_nullable(), is_instant(), instant_value(). dict_col_t::remove_instant(): Remove the 'instant ADD' status of a column. dict_col_t::name(const dict_table_t& table): Replaces dict_table_get_col_name(). dict_index_t::n_core_fields: The original number of fields. For secondary indexes and if instant ADD COLUMN has not been used, this will be equal to dict_index_t::n_fields. dict_index_t::n_core_null_bytes: Number of bytes needed to represent the null flags; usually equal to UT_BITS_IN_BYTES(n_nullable). dict_index_t::NO_CORE_NULL_BYTES: Magic value signalling that n_core_null_bytes was not initialized yet from the clustered index root page. dict_index_t: Add the accessors is_instant(), is_clust(), get_n_nullable(), instant_field_value(). dict_index_t::instant_add_field(): Adjust clustered index metadata for instant ADD COLUMN. dict_index_t::remove_instant(): Remove the 'instant ADD' status of a clustered index when the table becomes empty, or the very first instant ADD COLUMN operation is rolled back. dict_table_t: Add the accessors is_instant(), is_temporary(), supports_instant(). dict_table_t::instant_add_column(): Adjust metadata for instant ADD COLUMN. dict_table_t::rollback_instant(): Adjust metadata on the rollback of instant ADD COLUMN. prepare_inplace_alter_table_dict(): First create the ctx->new_table, and only then decide if the table really needs to be rebuilt. We must split the creation of table or index metadata from the creation of the dictionary table records and the creation of the data. In this way, we can transform a table-rebuilding operation into an instant ADD COLUMN operation. Dictionary objects will only be added to cache when table rebuilding or index creation is needed. The ctx->instant_table will never be added to cache. dict_table_t::add_to_cache(): Modified and renamed from dict_table_add_to_cache(). Do not modify the table metadata. Let the callers invoke dict_table_add_system_columns() and if needed, set can_be_evicted. dict_create_sys_tables_tuple(), dict_create_table_step(): Omit the system columns (which will now exist in the dict_table_t object already at this point). dict_create_table_step(): Expect the callers to invoke dict_table_add_system_columns(). pars_create_table(): Before creating the table creation execution graph, invoke dict_table_add_system_columns(). row_create_table_for_mysql(): Expect all callers to invoke dict_table_add_system_columns(). create_index_dict(): Replaces row_merge_create_index_graph(). innodb_update_n_cols(): Renamed from innobase_update_n_virtual(). Call my_error() if an error occurs. btr_cur_instant_init(), btr_cur_instant_init_low(), btr_cur_instant_root_init(): Load additional metadata from the clustered index and set dict_index_t::n_core_null_bytes. This is invoked when table metadata is first loaded into the data dictionary. dict_boot(): Initialize n_core_null_bytes for the four hard-coded dictionary tables. dict_create_index_step(): Initialize n_core_null_bytes. This is executed as part of CREATE TABLE. dict_index_build_internal_clust(): Initialize n_core_null_bytes to NO_CORE_NULL_BYTES if table->supports_instant(). row_create_index_for_mysql(): Initialize n_core_null_bytes for CREATE TEMPORARY TABLE. commit_cache_norebuild(): Call the code to rename or enlarge columns in the cache only if instant ADD COLUMN is not being used. (Instant ADD COLUMN would copy all column metadata from instant_table to old_table, including the names and lengths.) PAGE_INSTANT: A new 13-bit field for storing dict_index_t::n_core_fields. This is repurposing the 16-bit field PAGE_DIRECTION, of which only the least significant 3 bits were used. The original byte containing PAGE_DIRECTION will be accessible via the new constant PAGE_DIRECTION_B. page_get_instant(), page_set_instant(): Accessors for the PAGE_INSTANT. page_ptr_get_direction(), page_get_direction(), page_ptr_set_direction(): Accessors for PAGE_DIRECTION. page_direction_reset(): Reset PAGE_DIRECTION, PAGE_N_DIRECTION. page_direction_increment(): Increment PAGE_N_DIRECTION and set PAGE_DIRECTION. rec_get_offsets(): Use the 'leaf' parameter for non-debug purposes, and assume that heap_no is always set. Initialize all dict_index_t::n_fields for ROW_FORMAT=REDUNDANT records, even if the record contains fewer fields. rec_offs_make_valid(): Add the parameter 'leaf'. rec_copy_prefix_to_dtuple(): Assert that the tuple is only built on the core fields. Instant ADD COLUMN only applies to the clustered index, and we should never build a search key that has more than the PRIMARY KEY and possibly DB_TRX_ID,DB_ROLL_PTR. All these columns are always present. dict_index_build_data_tuple(): Remove assertions that would be duplicated in rec_copy_prefix_to_dtuple(). rec_init_offsets(): Support ROW_FORMAT=REDUNDANT records whose number of fields is between n_core_fields and n_fields. cmp_rec_rec_with_match(): Implement the comparison between two MIN_REC_FLAG records. trx_t::in_rollback: Make the field available in non-debug builds. trx_start_for_ddl_low(): Remove dangerous error-tolerance. A dictionary transaction must be flagged as such before it has generated any undo log records. This is because trx_undo_assign_undo() will mark the transaction as a dictionary transaction in the undo log header right before the very first undo log record is being written. btr_index_rec_validate(): Account for instant ADD COLUMN row_undo_ins_remove_clust_rec(): On the rollback of an insert into SYS_COLUMNS, revert instant ADD COLUMN in the cache by removing the last column from the table and the clustered index. row_search_on_row_ref(), row_undo_mod_parse_undo_rec(), row_undo_mod(), trx_undo_update_rec_get_update(): Handle the 'default row' as a special case. dtuple_t::trim(index): Omit a redundant suffix of an index tuple right before insert or update. After instant ADD COLUMN, if the last fields of a clustered index tuple match the 'default row', there is no need to store them. While trimming the entry, we must hold a page latch, so that the table cannot be emptied and the 'default row' be deleted. btr_cur_optimistic_update(), btr_cur_pessimistic_update(), row_upd_clust_rec_by_insert(), row_ins_clust_index_entry_low(): Invoke dtuple_t::trim() if needed. row_ins_clust_index_entry(): Restore dtuple_t::n_fields after calling row_ins_clust_index_entry_low(). rec_get_converted_size(), rec_get_converted_size_comp(): Allow the number of fields to be between n_core_fields and n_fields. Do not support infimum,supremum. They are never supposed to be stored in dtuple_t, because page creation nowadays uses a lower-level method for initializing them. rec_convert_dtuple_to_rec_comp(): Assign the status bits based on the number of fields. btr_cur_trim(): In an update, trim the index entry as needed. For the 'default row', handle rollback specially. For user records, omit fields that match the 'default row'. btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete(): Skip locking and adaptive hash index for the 'default row'. row_log_table_apply_convert_mrec(): Replace 'default row' values if needed. In the temporary file that is applied by row_log_table_apply(), we must identify whether the records contain the extra header for instantly added columns. For now, we will allocate an additional byte for this for ROW_T_INSERT and ROW_T_UPDATE records when the source table has been subject to instant ADD COLUMN. The ROW_T_DELETE records are fine, as they will be converted and will only contain 'core' columns (PRIMARY KEY and some system columns) that are converted from dtuple_t. rec_get_converted_size_temp(), rec_init_offsets_temp(), rec_convert_dtuple_to_temp(): Add the parameter 'status'. REC_INFO_DEFAULT_ROW = REC_INFO_MIN_REC_FLAG | REC_STATUS_COLUMNS_ADDED: An info_bits constant for distinguishing the 'default row' record. rec_comp_status_t: An enum of the status bit values. rec_leaf_format: An enum that replaces the bool parameter of rec_init_offsets_comp_ordinary().
8 years ago
MDEV-24971 InnoDB access freed virtual column after rollback of secondary index Problem: ======== InnoDB fails to clean the index stub if it fails to add the virtual index which contains new virtual column. But it clears the newly virtual column from index in clear_added_indexes() during inplace_alter_table. On commit, InnoDB evicts and reload the table. In case of rollback, it doesn't happen. InnoDB clears the ABORTED index while opening the table or doing the DDL. In the mean time, InnoDB can access the dropped virtual index columns while creating prebuilt or rollback of concurrent DML. Solution: ========== (1) InnoDB should maintain newly added virtual column while rollbacking the newly added virtual index. (2) InnoDB must not defer the index removal if the alter table is executed with LOCK=EXCLUSIVE. (3) For LOCK=SHARED, InnoDB should check whether the table has any other transaction lock other than alter transaction before deferring the index stub. Replaced has_new_v_col with dict_add_vcol_info in dict_index_t to indicate whether the index has any new virtual column. dict_index_t::has_new_v_col(): Returns whether the index has newly added virtual column, it doesn't say which columns are newly added virtual column ha_innobase_inplace_ctx::is_new_vcol(): Return whether the given column is added as a part of the current alter. ha_innobase_inplace_ctx::clean_new_vcol_index(): Copy the newly added virtual column to new_vcol_info in dict_index_t. Replace the column in the index fields with virtual column stored in new_vcol_info. dict_index_t::assign_new_v_col(): Store the number of virtual column added in index as a part of alter table. dict_index_t::get_n_new_vcol(): Get the number of newly added virtual column dict_index_t::assign_drop_v_col(): Allocate the memory for adding new virtual column in new_vcol_info. dict_index_t::add_drop_v_col(): Add the newly added virtual column in new_vcol_info. dict_table_t::has_lock_for_other_trx(): Whether the table has any other transaction lock than given transaction. row_merge_drop_indexes(): Add parameter alter_trx and check whether the table has any other lock than alter transaction.
5 years ago
MDEV-12253: Buffer pool blocks are accessed after they have been freed Problem was that bpage was referenced after it was already freed from LRU. Fixed by adding a new variable encrypted that is passed down to buf_page_check_corrupt() and used in buf_page_get_gen() to stop processing page read. This patch should also address following test failures and bugs: MDEV-12419: IMPORT should not look up tablespace in PageConverter::validate(). This is now removed. MDEV-10099: encryption.innodb_onlinealter_encryption fails sporadically in buildbot MDEV-11420: encryption.innodb_encryption-page-compression failed in buildbot MDEV-11222: encryption.encrypt_and_grep failed in buildbot on P8 Removed dict_table_t::is_encrypted and dict_table_t::ibd_file_missing and replaced these with dict_table_t::file_unreadable. Table ibd file is missing if fil_get_space(space_id) returns NULL and encrypted if not. Removed dict_table_t::is_corrupted field. Ported FilSpace class from 10.2 and using that on buf_page_check_corrupt(), buf_page_decrypt_after_read(), buf_page_encrypt_before_write(), buf_dblwr_process(), buf_read_page(), dict_stats_save_defrag_stats(). Added test cases when enrypted page could be read while doing redo log crash recovery. Also added test case for row compressed blobs. btr_cur_open_at_index_side_func(), btr_cur_open_at_rnd_pos_func(): Avoid referencing block that is NULL. buf_page_get_zip(): Issue error if page read fails. buf_page_get_gen(): Use dberr_t for error detection and do not reference bpage after we hare freed it. buf_mark_space_corrupt(): remove bpage from LRU also when it is encrypted. buf_page_check_corrupt(): @return DB_SUCCESS if page has been read and is not corrupted, DB_PAGE_CORRUPTED if page based on checksum check is corrupted, DB_DECRYPTION_FAILED if page post encryption checksum matches but after decryption normal page checksum does not match. In read case only DB_SUCCESS is possible. buf_page_io_complete(): use dberr_t for error handling. buf_flush_write_block_low(), buf_read_ahead_random(), buf_read_page_async(), buf_read_ahead_linear(), buf_read_ibuf_merge_pages(), buf_read_recv_pages(), fil_aio_wait(): Issue error if page read fails. btr_pcur_move_to_next_page(): Do not reference page if it is NULL. Introduced dict_table_t::is_readable() and dict_index_t::is_readable() that will return true if tablespace exists and pages read from tablespace are not corrupted or page decryption failed. Removed buf_page_t::key_version. After page decryption the key version is not removed from page frame. For unencrypted pages, old key_version is removed at buf_page_encrypt_before_write() dict_stats_update_transient_for_index(), dict_stats_update_transient() Do not continue if table decryption failed or table is corrupted. dict0stats.cc: Introduced a dict_stats_report_error function to avoid code duplication. fil_parse_write_crypt_data(): Check that key read from redo log entry is found from encryption plugin and if it is not, refuse to start. PageConverter::validate(): Removed access to fil_space_t as tablespace is not available during import. Fixed error code on innodb.innodb test. Merged test cased innodb-bad-key-change5 and innodb-bad-key-shutdown to innodb-bad-key-change2. Removed innodb-bad-key-change5 test. Decreased unnecessary complexity on some long lasting tests. Removed fil_inc_pending_ops(), fil_decr_pending_ops(), fil_get_first_space(), fil_get_next_space(), fil_get_first_space_safe(), fil_get_next_space_safe() functions. fil_space_verify_crypt_checksum(): Fixed bug found using ASAN where FIL_PAGE_END_LSN_OLD_CHECKSUM field was incorrectly accessed from row compressed tables. Fixed out of page frame bug for row compressed tables in fil_space_verify_crypt_checksum() found using ASAN. Incorrect function was called for compressed table. Added new tests for discard, rename table and drop (we should allow them even when page decryption fails). Alter table rename is not allowed. Added test for restart with innodb-force-recovery=1 when page read on redo-recovery cant be decrypted. Added test for corrupted table where both page data and FIL_PAGE_FILE_FLUSH_LSN_OR_KEY_VERSION is corrupted. Adjusted the test case innodb_bug14147491 so that it does not anymore expect crash. Instead table is just mostly not usable. fil0fil.h: fil_space_acquire_low is not visible function and fil_space_acquire and fil_space_acquire_silent are inline functions. FilSpace class uses fil_space_acquire_low directly. recv_apply_hashed_log_recs() does not return anything.
9 years ago
Merge 10.1 into 10.2 This only merges MDEV-12253, adapting it to MDEV-12602 which is already present in 10.2 but not yet in the 10.1 revision that is being merged. TODO: Error handling in crash recovery needs to be improved. If a page cannot be decrypted (or read), we should cleanly abort the startup. If innodb_force_recovery is specified, we should ignore the problematic page and apply redo log to other pages. Currently, the test encryption.innodb-redo-badkey randomly fails like this (the last messages are from cmake -DWITH_ASAN): 2017-05-05 10:19:40 140037071685504 [Note] InnoDB: Starting crash recovery from checkpoint LSN=1635994 2017-05-05 10:19:40 140037071685504 [ERROR] InnoDB: Missing MLOG_FILE_NAME or MLOG_FILE_DELETE before MLOG_CHECKPOINT for tablespace 1 2017-05-05 10:19:40 140037071685504 [ERROR] InnoDB: Plugin initialization aborted at srv0start.cc[2201] with error Data structure corruption 2017-05-05 10:19:41 140037071685504 [Note] InnoDB: Starting shutdown... i================================================================= ==5226==ERROR: AddressSanitizer: attempting free on address which was not malloc()-ed: 0x612000018588 in thread T0 #0 0x736750 in operator delete(void*) (/mariadb/server/build/sql/mysqld+0x736750) #1 0x1e4833f in LatchCounter::~LatchCounter() /mariadb/server/storage/innobase/include/sync0types.h:599:4 #2 0x1e480b8 in LatchMeta<LatchCounter>::~LatchMeta() /mariadb/server/storage/innobase/include/sync0types.h:786:17 #3 0x1e35509 in sync_latch_meta_destroy() /mariadb/server/storage/innobase/sync/sync0debug.cc:1622:3 #4 0x1e35314 in sync_check_close() /mariadb/server/storage/innobase/sync/sync0debug.cc:1839:2 #5 0x1dfdc18 in innodb_shutdown() /mariadb/server/storage/innobase/srv/srv0start.cc:2888:2 #6 0x197e5e6 in innobase_init(void*) /mariadb/server/storage/innobase/handler/ha_innodb.cc:4475:3
9 years ago
MDEV-11369 Instant ADD COLUMN for InnoDB For InnoDB tables, adding, dropping and reordering columns has required a rebuild of the table and all its indexes. Since MySQL 5.6 (and MariaDB 10.0) this has been supported online (LOCK=NONE), allowing concurrent modification of the tables. This work revises the InnoDB ROW_FORMAT=REDUNDANT, ROW_FORMAT=COMPACT and ROW_FORMAT=DYNAMIC so that columns can be appended instantaneously, with only minor changes performed to the table structure. The counter innodb_instant_alter_column in INFORMATION_SCHEMA.GLOBAL_STATUS is incremented whenever a table rebuild operation is converted into an instant ADD COLUMN operation. ROW_FORMAT=COMPRESSED tables will not support instant ADD COLUMN. Some usability limitations will be addressed in subsequent work: MDEV-13134 Introduce ALTER TABLE attributes ALGORITHM=NOCOPY and ALGORITHM=INSTANT MDEV-14016 Allow instant ADD COLUMN, ADD INDEX, LOCK=NONE The format of the clustered index (PRIMARY KEY) is changed as follows: (1) The FIL_PAGE_TYPE of the root page will be FIL_PAGE_TYPE_INSTANT, and a new field PAGE_INSTANT will contain the original number of fields in the clustered index ('core' fields). If instant ADD COLUMN has not been used or the table becomes empty, or the very first instant ADD COLUMN operation is rolled back, the fields PAGE_INSTANT and FIL_PAGE_TYPE will be reset to 0 and FIL_PAGE_INDEX. (2) A special 'default row' record is inserted into the leftmost leaf, between the page infimum and the first user record. This record is distinguished by the REC_INFO_MIN_REC_FLAG, and it is otherwise in the same format as records that contain values for the instantly added columns. This 'default row' always has the same number of fields as the clustered index according to the table definition. The values of 'core' fields are to be ignored. For other fields, the 'default row' will contain the default values as they were during the ALTER TABLE statement. (If the column default values are changed later, those values will only be stored in the .frm file. The 'default row' will contain the original evaluated values, which must be the same for every row.) The 'default row' must be completely hidden from higher-level access routines. Assertions have been added to ensure that no 'default row' is ever present in the adaptive hash index or in locked records. The 'default row' is never delete-marked. (3) In clustered index leaf page records, the number of fields must reside between the number of 'core' fields (dict_index_t::n_core_fields introduced in this work) and dict_index_t::n_fields. If the number of fields is less than dict_index_t::n_fields, the missing fields are replaced with the column value of the 'default row'. Note: The number of fields in the record may shrink if some of the last instantly added columns are updated to the value that is in the 'default row'. The function btr_cur_trim() implements this 'compression' on update and rollback; dtuple::trim() implements it on insert. (4) In ROW_FORMAT=COMPACT and ROW_FORMAT=DYNAMIC records, the new status value REC_STATUS_COLUMNS_ADDED will indicate the presence of a new record header that will encode n_fields-n_core_fields-1 in 1 or 2 bytes. (In ROW_FORMAT=REDUNDANT records, the record header always explicitly encodes the number of fields.) We introduce the undo log record type TRX_UNDO_INSERT_DEFAULT for covering the insert of the 'default row' record when instant ADD COLUMN is used for the first time. Subsequent instant ADD COLUMN can use TRX_UNDO_UPD_EXIST_REC. This is joint work with Vin Chen (陈福荣) from Tencent. The design that was discussed in April 2017 would not have allowed import or export of data files, because instead of the 'default row' it would have introduced a data dictionary table. The test rpl.rpl_alter_instant is exactly as contributed in pull request #408. The test innodb.instant_alter is based on a contributed test. The redo log record format changes for ROW_FORMAT=DYNAMIC and ROW_FORMAT=COMPACT are as contributed. (With this change present, crash recovery from MariaDB 10.3.1 will fail in spectacular ways!) Also the semantics of higher-level redo log records that modify the PAGE_INSTANT field is changed. The redo log format version identifier was already changed to LOG_HEADER_FORMAT_CURRENT=103 in MariaDB 10.3.1. Everything else has been rewritten by me. Thanks to Elena Stepanova, the code has been tested extensively. When rolling back an instant ADD COLUMN operation, we must empty the PAGE_FREE list after deleting or shortening the 'default row' record, by calling either btr_page_empty() or btr_page_reorganize(). We must know the size of each entry in the PAGE_FREE list. If rollback left a freed copy of the 'default row' in the PAGE_FREE list, we would be unable to determine its size (if it is in ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC) because it would contain more fields than the rolled-back definition of the clustered index. UNIV_SQL_DEFAULT: A new special constant that designates an instantly added column that is not present in the clustered index record. len_is_stored(): Check if a length is an actual length. There are two magic length values: UNIV_SQL_DEFAULT, UNIV_SQL_NULL. dict_col_t::def_val: The 'default row' value of the column. If the column is not added instantly, def_val.len will be UNIV_SQL_DEFAULT. dict_col_t: Add the accessors is_virtual(), is_nullable(), is_instant(), instant_value(). dict_col_t::remove_instant(): Remove the 'instant ADD' status of a column. dict_col_t::name(const dict_table_t& table): Replaces dict_table_get_col_name(). dict_index_t::n_core_fields: The original number of fields. For secondary indexes and if instant ADD COLUMN has not been used, this will be equal to dict_index_t::n_fields. dict_index_t::n_core_null_bytes: Number of bytes needed to represent the null flags; usually equal to UT_BITS_IN_BYTES(n_nullable). dict_index_t::NO_CORE_NULL_BYTES: Magic value signalling that n_core_null_bytes was not initialized yet from the clustered index root page. dict_index_t: Add the accessors is_instant(), is_clust(), get_n_nullable(), instant_field_value(). dict_index_t::instant_add_field(): Adjust clustered index metadata for instant ADD COLUMN. dict_index_t::remove_instant(): Remove the 'instant ADD' status of a clustered index when the table becomes empty, or the very first instant ADD COLUMN operation is rolled back. dict_table_t: Add the accessors is_instant(), is_temporary(), supports_instant(). dict_table_t::instant_add_column(): Adjust metadata for instant ADD COLUMN. dict_table_t::rollback_instant(): Adjust metadata on the rollback of instant ADD COLUMN. prepare_inplace_alter_table_dict(): First create the ctx->new_table, and only then decide if the table really needs to be rebuilt. We must split the creation of table or index metadata from the creation of the dictionary table records and the creation of the data. In this way, we can transform a table-rebuilding operation into an instant ADD COLUMN operation. Dictionary objects will only be added to cache when table rebuilding or index creation is needed. The ctx->instant_table will never be added to cache. dict_table_t::add_to_cache(): Modified and renamed from dict_table_add_to_cache(). Do not modify the table metadata. Let the callers invoke dict_table_add_system_columns() and if needed, set can_be_evicted. dict_create_sys_tables_tuple(), dict_create_table_step(): Omit the system columns (which will now exist in the dict_table_t object already at this point). dict_create_table_step(): Expect the callers to invoke dict_table_add_system_columns(). pars_create_table(): Before creating the table creation execution graph, invoke dict_table_add_system_columns(). row_create_table_for_mysql(): Expect all callers to invoke dict_table_add_system_columns(). create_index_dict(): Replaces row_merge_create_index_graph(). innodb_update_n_cols(): Renamed from innobase_update_n_virtual(). Call my_error() if an error occurs. btr_cur_instant_init(), btr_cur_instant_init_low(), btr_cur_instant_root_init(): Load additional metadata from the clustered index and set dict_index_t::n_core_null_bytes. This is invoked when table metadata is first loaded into the data dictionary. dict_boot(): Initialize n_core_null_bytes for the four hard-coded dictionary tables. dict_create_index_step(): Initialize n_core_null_bytes. This is executed as part of CREATE TABLE. dict_index_build_internal_clust(): Initialize n_core_null_bytes to NO_CORE_NULL_BYTES if table->supports_instant(). row_create_index_for_mysql(): Initialize n_core_null_bytes for CREATE TEMPORARY TABLE. commit_cache_norebuild(): Call the code to rename or enlarge columns in the cache only if instant ADD COLUMN is not being used. (Instant ADD COLUMN would copy all column metadata from instant_table to old_table, including the names and lengths.) PAGE_INSTANT: A new 13-bit field for storing dict_index_t::n_core_fields. This is repurposing the 16-bit field PAGE_DIRECTION, of which only the least significant 3 bits were used. The original byte containing PAGE_DIRECTION will be accessible via the new constant PAGE_DIRECTION_B. page_get_instant(), page_set_instant(): Accessors for the PAGE_INSTANT. page_ptr_get_direction(), page_get_direction(), page_ptr_set_direction(): Accessors for PAGE_DIRECTION. page_direction_reset(): Reset PAGE_DIRECTION, PAGE_N_DIRECTION. page_direction_increment(): Increment PAGE_N_DIRECTION and set PAGE_DIRECTION. rec_get_offsets(): Use the 'leaf' parameter for non-debug purposes, and assume that heap_no is always set. Initialize all dict_index_t::n_fields for ROW_FORMAT=REDUNDANT records, even if the record contains fewer fields. rec_offs_make_valid(): Add the parameter 'leaf'. rec_copy_prefix_to_dtuple(): Assert that the tuple is only built on the core fields. Instant ADD COLUMN only applies to the clustered index, and we should never build a search key that has more than the PRIMARY KEY and possibly DB_TRX_ID,DB_ROLL_PTR. All these columns are always present. dict_index_build_data_tuple(): Remove assertions that would be duplicated in rec_copy_prefix_to_dtuple(). rec_init_offsets(): Support ROW_FORMAT=REDUNDANT records whose number of fields is between n_core_fields and n_fields. cmp_rec_rec_with_match(): Implement the comparison between two MIN_REC_FLAG records. trx_t::in_rollback: Make the field available in non-debug builds. trx_start_for_ddl_low(): Remove dangerous error-tolerance. A dictionary transaction must be flagged as such before it has generated any undo log records. This is because trx_undo_assign_undo() will mark the transaction as a dictionary transaction in the undo log header right before the very first undo log record is being written. btr_index_rec_validate(): Account for instant ADD COLUMN row_undo_ins_remove_clust_rec(): On the rollback of an insert into SYS_COLUMNS, revert instant ADD COLUMN in the cache by removing the last column from the table and the clustered index. row_search_on_row_ref(), row_undo_mod_parse_undo_rec(), row_undo_mod(), trx_undo_update_rec_get_update(): Handle the 'default row' as a special case. dtuple_t::trim(index): Omit a redundant suffix of an index tuple right before insert or update. After instant ADD COLUMN, if the last fields of a clustered index tuple match the 'default row', there is no need to store them. While trimming the entry, we must hold a page latch, so that the table cannot be emptied and the 'default row' be deleted. btr_cur_optimistic_update(), btr_cur_pessimistic_update(), row_upd_clust_rec_by_insert(), row_ins_clust_index_entry_low(): Invoke dtuple_t::trim() if needed. row_ins_clust_index_entry(): Restore dtuple_t::n_fields after calling row_ins_clust_index_entry_low(). rec_get_converted_size(), rec_get_converted_size_comp(): Allow the number of fields to be between n_core_fields and n_fields. Do not support infimum,supremum. They are never supposed to be stored in dtuple_t, because page creation nowadays uses a lower-level method for initializing them. rec_convert_dtuple_to_rec_comp(): Assign the status bits based on the number of fields. btr_cur_trim(): In an update, trim the index entry as needed. For the 'default row', handle rollback specially. For user records, omit fields that match the 'default row'. btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete(): Skip locking and adaptive hash index for the 'default row'. row_log_table_apply_convert_mrec(): Replace 'default row' values if needed. In the temporary file that is applied by row_log_table_apply(), we must identify whether the records contain the extra header for instantly added columns. For now, we will allocate an additional byte for this for ROW_T_INSERT and ROW_T_UPDATE records when the source table has been subject to instant ADD COLUMN. The ROW_T_DELETE records are fine, as they will be converted and will only contain 'core' columns (PRIMARY KEY and some system columns) that are converted from dtuple_t. rec_get_converted_size_temp(), rec_init_offsets_temp(), rec_convert_dtuple_to_temp(): Add the parameter 'status'. REC_INFO_DEFAULT_ROW = REC_INFO_MIN_REC_FLAG | REC_STATUS_COLUMNS_ADDED: An info_bits constant for distinguishing the 'default row' record. rec_comp_status_t: An enum of the status bit values. rec_leaf_format: An enum that replaces the bool parameter of rec_init_offsets_comp_ordinary().
8 years ago
MDEV-15662 Instant DROP COLUMN or changing the order of columns Allow ADD COLUMN anywhere in a table, not only adding as the last column. Allow instant DROP COLUMN and instant changing the order of columns. The added columns will always be added last in clustered index records. In new records, instantly dropped columns will be stored as NULL or empty when possible. Information about dropped and reordered columns will be written in a metadata BLOB (mblob), which is stored before the first 'user' field in the hidden metadata record at the start of the clustered index. The presence of mblob is indicated by setting the delete-mark flag in the metadata record. The metadata BLOB stores the number of clustered index fields, followed by an array of column information for each field. For dropped columns, we store the NOT NULL flag, the fixed length, and for variable-length columns, whether the maximum length exceeded 255 bytes. For non-dropped columns, we store the column position. Unlike with MDEV-11369, when a table becomes empty, it cannot be converted back to the canonical format. The reason for this is that other threads may hold cached objects such as row_prebuilt_t::ins_node that could refer to dropped or reordered index fields. For instant DROP COLUMN and ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC, we must store the n_core_null_bytes in the root page, so that the chain of node pointer records can be followed in order to reach the leftmost leaf page where the metadata record is located. If the mblob is present, we will zero-initialize the strings "infimum" and "supremum" in the root page, and use the last byte of "supremum" for storing the number of null bytes (which are allocated but useless on node pointer pages). This is necessary for btr_cur_instant_init_metadata() to be able to navigate to the mblob. If the PRIMARY KEY contains any variable-length column and some nullable columns were instantly dropped, the dict_index_t::n_nullable in the data dictionary could be smaller than it actually is in the non-leaf pages. Because of this, the non-leaf pages could use more bytes for the null flags than the data dictionary expects, and we could be reading the lengths of the variable-length columns from the wrong offset, and thus reading the child page number from wrong place. This is the result of two design mistakes that involve unnecessary storage of data: First, it is nonsense to store any data fields for the leftmost node pointer records, because the comparisons would be resolved by the MIN_REC_FLAG alone. Second, there cannot be any null fields in the clustered index node pointer fields, but we nevertheless reserve space for all the null flags. Limitations (future work): MDEV-17459 Allow instant ALTER TABLE even if FULLTEXT INDEX exists MDEV-17468 Avoid table rebuild on operations on generated columns MDEV-17494 Refuse ALGORITHM=INSTANT when the row size is too large btr_page_reorganize_low(): Preserve any metadata in the root page. Call lock_move_reorganize_page() only after restoring the "infimum" and "supremum" records, to avoid a memcmp() assertion failure. dict_col_t::DROPPED: Magic value for dict_col_t::ind. dict_col_t::clear_instant(): Renamed from dict_col_t::remove_instant(). Do not assert that the column was instantly added, because we sometimes call this unconditionally for all columns. Convert an instantly added column to a "core column". The old name remove_instant() could be mistaken to refer to "instant DROP COLUMN". dict_col_t::is_added(): Rename from dict_col_t::is_instant(). dtype_t::metadata_blob_init(): Initialize the mblob data type. dtuple_t::is_metadata(), dtuple_t::is_alter_metadata(), upd_t::is_metadata(), upd_t::is_alter_metadata(): Check if info_bits refer to a metadata record. dict_table_t::instant: Metadata about dropped or reordered columns. dict_table_t::prepare_instant(): Prepare ha_innobase_inplace_ctx::instant_table for instant ALTER TABLE. innobase_instant_try() will pass this to dict_table_t::instant_column(). On rollback, dict_table_t::rollback_instant() will be called. dict_table_t::instant_column(): Renamed from instant_add_column(). Add the parameter col_map so that columns can be reordered. Copy and adjust v_cols[] as well. dict_table_t::find(): Find an old column based on a new column number. dict_table_t::serialise_columns(), dict_table_t::deserialise_columns(): Convert the mblob. dict_index_t::instant_metadata(): Create the metadata record for instant ALTER TABLE. Invoke dict_table_t::serialise_columns(). dict_index_t::reconstruct_fields(): Invoked by dict_table_t::deserialise_columns(). dict_index_t::clear_instant_alter(): Move the fields for the dropped columns to the end, and sort the surviving index fields in ascending order of column position. ha_innobase::check_if_supported_inplace_alter(): Do not allow adding a FTS_DOC_ID column if a hidden FTS_DOC_ID column exists due to FULLTEXT INDEX. (This always required ALGORITHM=COPY.) instant_alter_column_possible(): Add a parameter for InnoDB table, to check for additional conditions, such as the maximum number of index fields. ha_innobase_inplace_ctx::first_alter_pos: The first column whose position is affected by instant ADD, DROP, or changing the order of columns. innobase_build_col_map(): Skip added virtual columns. prepare_inplace_add_virtual(): Correctly compute num_to_add_vcol. Remove some unnecessary code. Note that the call to innodb_base_col_setup() should be executed later. commit_try_norebuild(): If ctx->is_instant(), let the virtual columns be added or dropped by innobase_instant_try(). innobase_instant_try(): Fill in a zero default value for the hidden column FTS_DOC_ID (to reduce the work needed in MDEV-17459). If any columns were dropped or reordered (or added not last), delete any SYS_COLUMNS records for the following columns, and insert SYS_COLUMNS records for all subsequent stored columns as well as for all virtual columns. If any virtual column is dropped, rewrite all virtual column metadata. Use a shortcut only for adding virtual columns. This is because innobase_drop_virtual_try() assumes that the dropped virtual columns still exist in ctx->old_table. innodb_update_cols(): Renamed from innodb_update_n_cols(). innobase_add_one_virtual(), innobase_insert_sys_virtual(): Change the return type to bool, and invoke my_error() when detecting an error. innodb_insert_sys_columns(): Insert a record into SYS_COLUMNS. Refactored from innobase_add_one_virtual() and innobase_instant_add_col(). innobase_instant_add_col(): Replace the parameter dfield with type. innobase_instant_drop_cols(): Drop matching columns from SYS_COLUMNS and all columns from SYS_VIRTUAL. innobase_add_virtual_try(), innobase_drop_virtual_try(): Let the caller invoke innodb_update_cols(). innobase_rename_column_try(): Skip dropped columns. commit_cache_norebuild(): Update table->fts->doc_col. dict_mem_table_col_rename_low(): Skip dropped columns. trx_undo_rec_get_partial_row(): Skip dropped columns. trx_undo_update_rec_get_update(): Handle the metadata BLOB correctly. trx_undo_page_report_modify(): Avoid out-of-bounds access to record fields. Log metadata records consistently. Apparently, the first fields of a clustered index may be updated in an update_undo vector when the index is ID_IND of SYS_FOREIGN, as part of renaming the table during ALTER TABLE. Normally, updates of the PRIMARY KEY should be logged as delete-mark and an insert. row_undo_mod_parse_undo_rec(), row_purge_parse_undo_rec(): Use trx_undo_metadata. row_undo_mod_clust_low(): On metadata rollback, roll back the root page too. row_undo_mod_clust(): Relax an assertion. The delete-mark flag was repurposed for ALTER TABLE metadata records. row_rec_to_index_entry_impl(): Add the template parameter mblob and the optional parameter info_bits for specifying the desired new info bits. For the metadata tuple, allow conversion between the original format (ADD COLUMN only) and the generic format (with hidden BLOB). Add the optional parameter "pad" to determine whether the tuple should be padded to the index fields (on ALTER TABLE it should), or whether it should remain at its original size (on rollback). row_build_index_entry_low(): Clean up the code, removing redundant variables and conditions. For instantly dropped columns, generate a dummy value that is NULL, the empty string, or a fixed length of NUL bytes, depending on the type of the dropped column. row_upd_clust_rec_by_insert_inherit_func(): On the update of PRIMARY KEY of a record that contained a dropped column whose value was stored externally, we will be inserting a dummy NULL or empty string value to the field of the dropped column. The externally stored column would eventually be dropped when purge removes the delete-marked record for the old PRIMARY KEY value. btr_index_rec_validate(): Recognize the metadata record. btr_discard_only_page_on_level(): Preserve the generic instant ALTER TABLE metadata. btr_set_instant(): Replaces page_set_instant(). This sets a clustered index root page to the appropriate format, or upgrades from the MDEV-11369 instant ADD COLUMN to generic ALTER TABLE format. btr_cur_instant_init_low(): Read and validate the metadata BLOB page before reconstructing the dictionary information based on it. btr_cur_instant_init_metadata(): Do not read any lengths from the metadata record header before reading the BLOB. At this point, we would not actually know how many nullable fields the metadata record contains. btr_cur_instant_root_init(): Initialize n_core_null_bytes in one of two possible ways. btr_cur_trim(): Handle the mblob record. row_metadata_to_tuple(): Convert a metadata record to a data tuple, based on the new info_bits of the metadata record. btr_cur_pessimistic_update(): Invoke row_metadata_to_tuple() if needed. Invoke dtuple_convert_big_rec() for metadata records if the record is too large, or if the mblob is not yet marked as externally stored. btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete(): When the last user record is deleted, do not delete the generic instant ALTER TABLE metadata record. Only delete MDEV-11369 instant ADD COLUMN metadata records. btr_cur_optimistic_insert(): Avoid unnecessary computation of rec_size. btr_pcur_store_position(): Allow a logically empty page to contain a metadata record for generic ALTER TABLE. REC_INFO_DEFAULT_ROW_ADD: Renamed from REC_INFO_DEFAULT_ROW. This is for the old instant ADD COLUMN (MDEV-11369) only. REC_INFO_DEFAULT_ROW_ALTER: The more generic metadata record, with additional information for dropped or reordered columns. rec_info_bits_valid(): Remove. The only case when this would fail is when the record is the generic ALTER TABLE metadata record. rec_is_alter_metadata(): Check if a record is the metadata record for instant ALTER TABLE (other than ADD COLUMN). NOTE: This function must not be invoked on node pointer records, because the delete-mark flag in those records may be set (it is garbage), and then a debug assertion could fail because index->is_instant() does not necessarily hold. rec_is_add_metadata(): Check if a record is MDEV-11369 ADD COLUMN metadata record (not more generic instant ALTER TABLE). rec_get_converted_size_comp_prefix_low(): Assume that the metadata field will be stored externally. In dtuple_convert_big_rec() during the rec_get_converted_size() call, it would not be there yet. rec_get_converted_size_comp(): Replace status,fields,n_fields with tuple. rec_init_offsets_comp_ordinary(), rec_get_converted_size_comp_prefix_low(), rec_convert_dtuple_to_rec_comp(): Add template<bool mblob = false>. With mblob=true, process a record with a metadata BLOB. rec_copy_prefix_to_buf(): Assert that no fields beyond the key and system columns are being copied. Exclude the metadata BLOB field. rec_convert_dtuple_to_metadata_comp(): Convert an alter metadata tuple into a record. row_upd_index_replace_metadata(): Apply an update vector to an alter_metadata tuple. row_log_allocate(): Replace dict_index_t::is_instant() with a more appropriate condition that ignores dict_table_t::instant. Only a table on which the MDEV-11369 ADD COLUMN was performed can "lose its instantness" when it becomes empty. After instant DROP COLUMN or reordering columns, we cannot simply convert the table to the canonical format, because the data dictionary cache and all possibly existing references to it from other client connection threads would have to be adjusted. row_quiesce_write_index_fields(): Do not crash when the table contains an instantly dropped column. Thanks to Thirunarayanan Balathandayuthapani for discussing the design and implementing an initial prototype of this. Thanks to Matthias Leich for testing.
7 years ago
MDEV-11369 Instant ADD COLUMN for InnoDB For InnoDB tables, adding, dropping and reordering columns has required a rebuild of the table and all its indexes. Since MySQL 5.6 (and MariaDB 10.0) this has been supported online (LOCK=NONE), allowing concurrent modification of the tables. This work revises the InnoDB ROW_FORMAT=REDUNDANT, ROW_FORMAT=COMPACT and ROW_FORMAT=DYNAMIC so that columns can be appended instantaneously, with only minor changes performed to the table structure. The counter innodb_instant_alter_column in INFORMATION_SCHEMA.GLOBAL_STATUS is incremented whenever a table rebuild operation is converted into an instant ADD COLUMN operation. ROW_FORMAT=COMPRESSED tables will not support instant ADD COLUMN. Some usability limitations will be addressed in subsequent work: MDEV-13134 Introduce ALTER TABLE attributes ALGORITHM=NOCOPY and ALGORITHM=INSTANT MDEV-14016 Allow instant ADD COLUMN, ADD INDEX, LOCK=NONE The format of the clustered index (PRIMARY KEY) is changed as follows: (1) The FIL_PAGE_TYPE of the root page will be FIL_PAGE_TYPE_INSTANT, and a new field PAGE_INSTANT will contain the original number of fields in the clustered index ('core' fields). If instant ADD COLUMN has not been used or the table becomes empty, or the very first instant ADD COLUMN operation is rolled back, the fields PAGE_INSTANT and FIL_PAGE_TYPE will be reset to 0 and FIL_PAGE_INDEX. (2) A special 'default row' record is inserted into the leftmost leaf, between the page infimum and the first user record. This record is distinguished by the REC_INFO_MIN_REC_FLAG, and it is otherwise in the same format as records that contain values for the instantly added columns. This 'default row' always has the same number of fields as the clustered index according to the table definition. The values of 'core' fields are to be ignored. For other fields, the 'default row' will contain the default values as they were during the ALTER TABLE statement. (If the column default values are changed later, those values will only be stored in the .frm file. The 'default row' will contain the original evaluated values, which must be the same for every row.) The 'default row' must be completely hidden from higher-level access routines. Assertions have been added to ensure that no 'default row' is ever present in the adaptive hash index or in locked records. The 'default row' is never delete-marked. (3) In clustered index leaf page records, the number of fields must reside between the number of 'core' fields (dict_index_t::n_core_fields introduced in this work) and dict_index_t::n_fields. If the number of fields is less than dict_index_t::n_fields, the missing fields are replaced with the column value of the 'default row'. Note: The number of fields in the record may shrink if some of the last instantly added columns are updated to the value that is in the 'default row'. The function btr_cur_trim() implements this 'compression' on update and rollback; dtuple::trim() implements it on insert. (4) In ROW_FORMAT=COMPACT and ROW_FORMAT=DYNAMIC records, the new status value REC_STATUS_COLUMNS_ADDED will indicate the presence of a new record header that will encode n_fields-n_core_fields-1 in 1 or 2 bytes. (In ROW_FORMAT=REDUNDANT records, the record header always explicitly encodes the number of fields.) We introduce the undo log record type TRX_UNDO_INSERT_DEFAULT for covering the insert of the 'default row' record when instant ADD COLUMN is used for the first time. Subsequent instant ADD COLUMN can use TRX_UNDO_UPD_EXIST_REC. This is joint work with Vin Chen (陈福荣) from Tencent. The design that was discussed in April 2017 would not have allowed import or export of data files, because instead of the 'default row' it would have introduced a data dictionary table. The test rpl.rpl_alter_instant is exactly as contributed in pull request #408. The test innodb.instant_alter is based on a contributed test. The redo log record format changes for ROW_FORMAT=DYNAMIC and ROW_FORMAT=COMPACT are as contributed. (With this change present, crash recovery from MariaDB 10.3.1 will fail in spectacular ways!) Also the semantics of higher-level redo log records that modify the PAGE_INSTANT field is changed. The redo log format version identifier was already changed to LOG_HEADER_FORMAT_CURRENT=103 in MariaDB 10.3.1. Everything else has been rewritten by me. Thanks to Elena Stepanova, the code has been tested extensively. When rolling back an instant ADD COLUMN operation, we must empty the PAGE_FREE list after deleting or shortening the 'default row' record, by calling either btr_page_empty() or btr_page_reorganize(). We must know the size of each entry in the PAGE_FREE list. If rollback left a freed copy of the 'default row' in the PAGE_FREE list, we would be unable to determine its size (if it is in ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC) because it would contain more fields than the rolled-back definition of the clustered index. UNIV_SQL_DEFAULT: A new special constant that designates an instantly added column that is not present in the clustered index record. len_is_stored(): Check if a length is an actual length. There are two magic length values: UNIV_SQL_DEFAULT, UNIV_SQL_NULL. dict_col_t::def_val: The 'default row' value of the column. If the column is not added instantly, def_val.len will be UNIV_SQL_DEFAULT. dict_col_t: Add the accessors is_virtual(), is_nullable(), is_instant(), instant_value(). dict_col_t::remove_instant(): Remove the 'instant ADD' status of a column. dict_col_t::name(const dict_table_t& table): Replaces dict_table_get_col_name(). dict_index_t::n_core_fields: The original number of fields. For secondary indexes and if instant ADD COLUMN has not been used, this will be equal to dict_index_t::n_fields. dict_index_t::n_core_null_bytes: Number of bytes needed to represent the null flags; usually equal to UT_BITS_IN_BYTES(n_nullable). dict_index_t::NO_CORE_NULL_BYTES: Magic value signalling that n_core_null_bytes was not initialized yet from the clustered index root page. dict_index_t: Add the accessors is_instant(), is_clust(), get_n_nullable(), instant_field_value(). dict_index_t::instant_add_field(): Adjust clustered index metadata for instant ADD COLUMN. dict_index_t::remove_instant(): Remove the 'instant ADD' status of a clustered index when the table becomes empty, or the very first instant ADD COLUMN operation is rolled back. dict_table_t: Add the accessors is_instant(), is_temporary(), supports_instant(). dict_table_t::instant_add_column(): Adjust metadata for instant ADD COLUMN. dict_table_t::rollback_instant(): Adjust metadata on the rollback of instant ADD COLUMN. prepare_inplace_alter_table_dict(): First create the ctx->new_table, and only then decide if the table really needs to be rebuilt. We must split the creation of table or index metadata from the creation of the dictionary table records and the creation of the data. In this way, we can transform a table-rebuilding operation into an instant ADD COLUMN operation. Dictionary objects will only be added to cache when table rebuilding or index creation is needed. The ctx->instant_table will never be added to cache. dict_table_t::add_to_cache(): Modified and renamed from dict_table_add_to_cache(). Do not modify the table metadata. Let the callers invoke dict_table_add_system_columns() and if needed, set can_be_evicted. dict_create_sys_tables_tuple(), dict_create_table_step(): Omit the system columns (which will now exist in the dict_table_t object already at this point). dict_create_table_step(): Expect the callers to invoke dict_table_add_system_columns(). pars_create_table(): Before creating the table creation execution graph, invoke dict_table_add_system_columns(). row_create_table_for_mysql(): Expect all callers to invoke dict_table_add_system_columns(). create_index_dict(): Replaces row_merge_create_index_graph(). innodb_update_n_cols(): Renamed from innobase_update_n_virtual(). Call my_error() if an error occurs. btr_cur_instant_init(), btr_cur_instant_init_low(), btr_cur_instant_root_init(): Load additional metadata from the clustered index and set dict_index_t::n_core_null_bytes. This is invoked when table metadata is first loaded into the data dictionary. dict_boot(): Initialize n_core_null_bytes for the four hard-coded dictionary tables. dict_create_index_step(): Initialize n_core_null_bytes. This is executed as part of CREATE TABLE. dict_index_build_internal_clust(): Initialize n_core_null_bytes to NO_CORE_NULL_BYTES if table->supports_instant(). row_create_index_for_mysql(): Initialize n_core_null_bytes for CREATE TEMPORARY TABLE. commit_cache_norebuild(): Call the code to rename or enlarge columns in the cache only if instant ADD COLUMN is not being used. (Instant ADD COLUMN would copy all column metadata from instant_table to old_table, including the names and lengths.) PAGE_INSTANT: A new 13-bit field for storing dict_index_t::n_core_fields. This is repurposing the 16-bit field PAGE_DIRECTION, of which only the least significant 3 bits were used. The original byte containing PAGE_DIRECTION will be accessible via the new constant PAGE_DIRECTION_B. page_get_instant(), page_set_instant(): Accessors for the PAGE_INSTANT. page_ptr_get_direction(), page_get_direction(), page_ptr_set_direction(): Accessors for PAGE_DIRECTION. page_direction_reset(): Reset PAGE_DIRECTION, PAGE_N_DIRECTION. page_direction_increment(): Increment PAGE_N_DIRECTION and set PAGE_DIRECTION. rec_get_offsets(): Use the 'leaf' parameter for non-debug purposes, and assume that heap_no is always set. Initialize all dict_index_t::n_fields for ROW_FORMAT=REDUNDANT records, even if the record contains fewer fields. rec_offs_make_valid(): Add the parameter 'leaf'. rec_copy_prefix_to_dtuple(): Assert that the tuple is only built on the core fields. Instant ADD COLUMN only applies to the clustered index, and we should never build a search key that has more than the PRIMARY KEY and possibly DB_TRX_ID,DB_ROLL_PTR. All these columns are always present. dict_index_build_data_tuple(): Remove assertions that would be duplicated in rec_copy_prefix_to_dtuple(). rec_init_offsets(): Support ROW_FORMAT=REDUNDANT records whose number of fields is between n_core_fields and n_fields. cmp_rec_rec_with_match(): Implement the comparison between two MIN_REC_FLAG records. trx_t::in_rollback: Make the field available in non-debug builds. trx_start_for_ddl_low(): Remove dangerous error-tolerance. A dictionary transaction must be flagged as such before it has generated any undo log records. This is because trx_undo_assign_undo() will mark the transaction as a dictionary transaction in the undo log header right before the very first undo log record is being written. btr_index_rec_validate(): Account for instant ADD COLUMN row_undo_ins_remove_clust_rec(): On the rollback of an insert into SYS_COLUMNS, revert instant ADD COLUMN in the cache by removing the last column from the table and the clustered index. row_search_on_row_ref(), row_undo_mod_parse_undo_rec(), row_undo_mod(), trx_undo_update_rec_get_update(): Handle the 'default row' as a special case. dtuple_t::trim(index): Omit a redundant suffix of an index tuple right before insert or update. After instant ADD COLUMN, if the last fields of a clustered index tuple match the 'default row', there is no need to store them. While trimming the entry, we must hold a page latch, so that the table cannot be emptied and the 'default row' be deleted. btr_cur_optimistic_update(), btr_cur_pessimistic_update(), row_upd_clust_rec_by_insert(), row_ins_clust_index_entry_low(): Invoke dtuple_t::trim() if needed. row_ins_clust_index_entry(): Restore dtuple_t::n_fields after calling row_ins_clust_index_entry_low(). rec_get_converted_size(), rec_get_converted_size_comp(): Allow the number of fields to be between n_core_fields and n_fields. Do not support infimum,supremum. They are never supposed to be stored in dtuple_t, because page creation nowadays uses a lower-level method for initializing them. rec_convert_dtuple_to_rec_comp(): Assign the status bits based on the number of fields. btr_cur_trim(): In an update, trim the index entry as needed. For the 'default row', handle rollback specially. For user records, omit fields that match the 'default row'. btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete(): Skip locking and adaptive hash index for the 'default row'. row_log_table_apply_convert_mrec(): Replace 'default row' values if needed. In the temporary file that is applied by row_log_table_apply(), we must identify whether the records contain the extra header for instantly added columns. For now, we will allocate an additional byte for this for ROW_T_INSERT and ROW_T_UPDATE records when the source table has been subject to instant ADD COLUMN. The ROW_T_DELETE records are fine, as they will be converted and will only contain 'core' columns (PRIMARY KEY and some system columns) that are converted from dtuple_t. rec_get_converted_size_temp(), rec_init_offsets_temp(), rec_convert_dtuple_to_temp(): Add the parameter 'status'. REC_INFO_DEFAULT_ROW = REC_INFO_MIN_REC_FLAG | REC_STATUS_COLUMNS_ADDED: An info_bits constant for distinguishing the 'default row' record. rec_comp_status_t: An enum of the status bit values. rec_leaf_format: An enum that replaces the bool parameter of rec_init_offsets_comp_ordinary().
8 years ago
MDEV-14407 Assertion failure during rollback Rollback attempted to dereference DB_ROLL_PTR=0, which cannot possibly be a valid undo log pointer. A safer canonical value would be roll_ptr_t(1) << ROLL_PTR_INSERT_FLAG_POS which is what was chosen in MDEV-12288, corresponding to reset_trx_id. No deterministic test case for the bug was found. The simplest test cases may be related to MDEV-11415, which suppresses undo logging for ALGORITHM=COPY operations. In those operations, in the spirit of MDEV-12288, we should actually have written reset_trx_id instead of using the transaction identifier of the current transaction (and a bogus value of DB_ROLL_PTR=0). However, thanks to MySQL Bug#28432 which I had fixed in MySQL 5.6.8 as part of WL#6255, access to the rebuilt table by earlier-started transactions should actually have been refused with ER_TABLE_DEF_CHANGED. reset_trx_id: Move the definition to data0type.cc and the declaration to data0type.h. btr_cur_ins_lock_and_undo(): When undo logging is disabled, use the safe value that corresponds to reset_trx_id. btr_cur_optimistic_insert(): Validate the DB_TRX_ID,DB_ROLL_PTR before inserting into a clustered index leaf page. ins_node_t::sys_buf[]: Replaces row_id_buf and trx_id_buf and some heap usage. row_ins_alloc_sys_fields(): Init ins_node_t::sys_buf[] to reset_trx_id. row_ins_buf(): Only if undo logging is enabled, copy trx->id to node->sys_buf. Otherwise, rely on the initialization in row_ins_alloc_sys_fields(). row_purge_reset_trx_id(): Invoke mlog_write_string() with reset_trx_id directly. (No functional change.) trx_undo_page_report_modify(): Assert that the DB_ROLL_PTR is not 0. trx_undo_get_undo_rec_low(): Assert that the roll_ptr is valid before trying to dereference it. dict_index_t::is_primary(): Check if the index is the primary key. PageConverter::adjust_cluster_record(): Fix MDEV-15249 Crash in MVCC read after IMPORT TABLESPACE by resetting the system fields to reset_trx_id instead of writing the current transaction ID (which will be committed at the end of the IMPORT TABLESPACE) and DB_ROLL_PTR=0. This can partially be viewed as a follow-up fix of MDEV-12288, because IMPORT should already then have written DB_TRX_ID=0 and DB_ROLL_PTR=1<<55 to prevent unnecessary DB_TRX_ID lookups in subsequent accesses to the table.
8 years ago
MDEV-11369 Instant ADD COLUMN for InnoDB For InnoDB tables, adding, dropping and reordering columns has required a rebuild of the table and all its indexes. Since MySQL 5.6 (and MariaDB 10.0) this has been supported online (LOCK=NONE), allowing concurrent modification of the tables. This work revises the InnoDB ROW_FORMAT=REDUNDANT, ROW_FORMAT=COMPACT and ROW_FORMAT=DYNAMIC so that columns can be appended instantaneously, with only minor changes performed to the table structure. The counter innodb_instant_alter_column in INFORMATION_SCHEMA.GLOBAL_STATUS is incremented whenever a table rebuild operation is converted into an instant ADD COLUMN operation. ROW_FORMAT=COMPRESSED tables will not support instant ADD COLUMN. Some usability limitations will be addressed in subsequent work: MDEV-13134 Introduce ALTER TABLE attributes ALGORITHM=NOCOPY and ALGORITHM=INSTANT MDEV-14016 Allow instant ADD COLUMN, ADD INDEX, LOCK=NONE The format of the clustered index (PRIMARY KEY) is changed as follows: (1) The FIL_PAGE_TYPE of the root page will be FIL_PAGE_TYPE_INSTANT, and a new field PAGE_INSTANT will contain the original number of fields in the clustered index ('core' fields). If instant ADD COLUMN has not been used or the table becomes empty, or the very first instant ADD COLUMN operation is rolled back, the fields PAGE_INSTANT and FIL_PAGE_TYPE will be reset to 0 and FIL_PAGE_INDEX. (2) A special 'default row' record is inserted into the leftmost leaf, between the page infimum and the first user record. This record is distinguished by the REC_INFO_MIN_REC_FLAG, and it is otherwise in the same format as records that contain values for the instantly added columns. This 'default row' always has the same number of fields as the clustered index according to the table definition. The values of 'core' fields are to be ignored. For other fields, the 'default row' will contain the default values as they were during the ALTER TABLE statement. (If the column default values are changed later, those values will only be stored in the .frm file. The 'default row' will contain the original evaluated values, which must be the same for every row.) The 'default row' must be completely hidden from higher-level access routines. Assertions have been added to ensure that no 'default row' is ever present in the adaptive hash index or in locked records. The 'default row' is never delete-marked. (3) In clustered index leaf page records, the number of fields must reside between the number of 'core' fields (dict_index_t::n_core_fields introduced in this work) and dict_index_t::n_fields. If the number of fields is less than dict_index_t::n_fields, the missing fields are replaced with the column value of the 'default row'. Note: The number of fields in the record may shrink if some of the last instantly added columns are updated to the value that is in the 'default row'. The function btr_cur_trim() implements this 'compression' on update and rollback; dtuple::trim() implements it on insert. (4) In ROW_FORMAT=COMPACT and ROW_FORMAT=DYNAMIC records, the new status value REC_STATUS_COLUMNS_ADDED will indicate the presence of a new record header that will encode n_fields-n_core_fields-1 in 1 or 2 bytes. (In ROW_FORMAT=REDUNDANT records, the record header always explicitly encodes the number of fields.) We introduce the undo log record type TRX_UNDO_INSERT_DEFAULT for covering the insert of the 'default row' record when instant ADD COLUMN is used for the first time. Subsequent instant ADD COLUMN can use TRX_UNDO_UPD_EXIST_REC. This is joint work with Vin Chen (陈福荣) from Tencent. The design that was discussed in April 2017 would not have allowed import or export of data files, because instead of the 'default row' it would have introduced a data dictionary table. The test rpl.rpl_alter_instant is exactly as contributed in pull request #408. The test innodb.instant_alter is based on a contributed test. The redo log record format changes for ROW_FORMAT=DYNAMIC and ROW_FORMAT=COMPACT are as contributed. (With this change present, crash recovery from MariaDB 10.3.1 will fail in spectacular ways!) Also the semantics of higher-level redo log records that modify the PAGE_INSTANT field is changed. The redo log format version identifier was already changed to LOG_HEADER_FORMAT_CURRENT=103 in MariaDB 10.3.1. Everything else has been rewritten by me. Thanks to Elena Stepanova, the code has been tested extensively. When rolling back an instant ADD COLUMN operation, we must empty the PAGE_FREE list after deleting or shortening the 'default row' record, by calling either btr_page_empty() or btr_page_reorganize(). We must know the size of each entry in the PAGE_FREE list. If rollback left a freed copy of the 'default row' in the PAGE_FREE list, we would be unable to determine its size (if it is in ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC) because it would contain more fields than the rolled-back definition of the clustered index. UNIV_SQL_DEFAULT: A new special constant that designates an instantly added column that is not present in the clustered index record. len_is_stored(): Check if a length is an actual length. There are two magic length values: UNIV_SQL_DEFAULT, UNIV_SQL_NULL. dict_col_t::def_val: The 'default row' value of the column. If the column is not added instantly, def_val.len will be UNIV_SQL_DEFAULT. dict_col_t: Add the accessors is_virtual(), is_nullable(), is_instant(), instant_value(). dict_col_t::remove_instant(): Remove the 'instant ADD' status of a column. dict_col_t::name(const dict_table_t& table): Replaces dict_table_get_col_name(). dict_index_t::n_core_fields: The original number of fields. For secondary indexes and if instant ADD COLUMN has not been used, this will be equal to dict_index_t::n_fields. dict_index_t::n_core_null_bytes: Number of bytes needed to represent the null flags; usually equal to UT_BITS_IN_BYTES(n_nullable). dict_index_t::NO_CORE_NULL_BYTES: Magic value signalling that n_core_null_bytes was not initialized yet from the clustered index root page. dict_index_t: Add the accessors is_instant(), is_clust(), get_n_nullable(), instant_field_value(). dict_index_t::instant_add_field(): Adjust clustered index metadata for instant ADD COLUMN. dict_index_t::remove_instant(): Remove the 'instant ADD' status of a clustered index when the table becomes empty, or the very first instant ADD COLUMN operation is rolled back. dict_table_t: Add the accessors is_instant(), is_temporary(), supports_instant(). dict_table_t::instant_add_column(): Adjust metadata for instant ADD COLUMN. dict_table_t::rollback_instant(): Adjust metadata on the rollback of instant ADD COLUMN. prepare_inplace_alter_table_dict(): First create the ctx->new_table, and only then decide if the table really needs to be rebuilt. We must split the creation of table or index metadata from the creation of the dictionary table records and the creation of the data. In this way, we can transform a table-rebuilding operation into an instant ADD COLUMN operation. Dictionary objects will only be added to cache when table rebuilding or index creation is needed. The ctx->instant_table will never be added to cache. dict_table_t::add_to_cache(): Modified and renamed from dict_table_add_to_cache(). Do not modify the table metadata. Let the callers invoke dict_table_add_system_columns() and if needed, set can_be_evicted. dict_create_sys_tables_tuple(), dict_create_table_step(): Omit the system columns (which will now exist in the dict_table_t object already at this point). dict_create_table_step(): Expect the callers to invoke dict_table_add_system_columns(). pars_create_table(): Before creating the table creation execution graph, invoke dict_table_add_system_columns(). row_create_table_for_mysql(): Expect all callers to invoke dict_table_add_system_columns(). create_index_dict(): Replaces row_merge_create_index_graph(). innodb_update_n_cols(): Renamed from innobase_update_n_virtual(). Call my_error() if an error occurs. btr_cur_instant_init(), btr_cur_instant_init_low(), btr_cur_instant_root_init(): Load additional metadata from the clustered index and set dict_index_t::n_core_null_bytes. This is invoked when table metadata is first loaded into the data dictionary. dict_boot(): Initialize n_core_null_bytes for the four hard-coded dictionary tables. dict_create_index_step(): Initialize n_core_null_bytes. This is executed as part of CREATE TABLE. dict_index_build_internal_clust(): Initialize n_core_null_bytes to NO_CORE_NULL_BYTES if table->supports_instant(). row_create_index_for_mysql(): Initialize n_core_null_bytes for CREATE TEMPORARY TABLE. commit_cache_norebuild(): Call the code to rename or enlarge columns in the cache only if instant ADD COLUMN is not being used. (Instant ADD COLUMN would copy all column metadata from instant_table to old_table, including the names and lengths.) PAGE_INSTANT: A new 13-bit field for storing dict_index_t::n_core_fields. This is repurposing the 16-bit field PAGE_DIRECTION, of which only the least significant 3 bits were used. The original byte containing PAGE_DIRECTION will be accessible via the new constant PAGE_DIRECTION_B. page_get_instant(), page_set_instant(): Accessors for the PAGE_INSTANT. page_ptr_get_direction(), page_get_direction(), page_ptr_set_direction(): Accessors for PAGE_DIRECTION. page_direction_reset(): Reset PAGE_DIRECTION, PAGE_N_DIRECTION. page_direction_increment(): Increment PAGE_N_DIRECTION and set PAGE_DIRECTION. rec_get_offsets(): Use the 'leaf' parameter for non-debug purposes, and assume that heap_no is always set. Initialize all dict_index_t::n_fields for ROW_FORMAT=REDUNDANT records, even if the record contains fewer fields. rec_offs_make_valid(): Add the parameter 'leaf'. rec_copy_prefix_to_dtuple(): Assert that the tuple is only built on the core fields. Instant ADD COLUMN only applies to the clustered index, and we should never build a search key that has more than the PRIMARY KEY and possibly DB_TRX_ID,DB_ROLL_PTR. All these columns are always present. dict_index_build_data_tuple(): Remove assertions that would be duplicated in rec_copy_prefix_to_dtuple(). rec_init_offsets(): Support ROW_FORMAT=REDUNDANT records whose number of fields is between n_core_fields and n_fields. cmp_rec_rec_with_match(): Implement the comparison between two MIN_REC_FLAG records. trx_t::in_rollback: Make the field available in non-debug builds. trx_start_for_ddl_low(): Remove dangerous error-tolerance. A dictionary transaction must be flagged as such before it has generated any undo log records. This is because trx_undo_assign_undo() will mark the transaction as a dictionary transaction in the undo log header right before the very first undo log record is being written. btr_index_rec_validate(): Account for instant ADD COLUMN row_undo_ins_remove_clust_rec(): On the rollback of an insert into SYS_COLUMNS, revert instant ADD COLUMN in the cache by removing the last column from the table and the clustered index. row_search_on_row_ref(), row_undo_mod_parse_undo_rec(), row_undo_mod(), trx_undo_update_rec_get_update(): Handle the 'default row' as a special case. dtuple_t::trim(index): Omit a redundant suffix of an index tuple right before insert or update. After instant ADD COLUMN, if the last fields of a clustered index tuple match the 'default row', there is no need to store them. While trimming the entry, we must hold a page latch, so that the table cannot be emptied and the 'default row' be deleted. btr_cur_optimistic_update(), btr_cur_pessimistic_update(), row_upd_clust_rec_by_insert(), row_ins_clust_index_entry_low(): Invoke dtuple_t::trim() if needed. row_ins_clust_index_entry(): Restore dtuple_t::n_fields after calling row_ins_clust_index_entry_low(). rec_get_converted_size(), rec_get_converted_size_comp(): Allow the number of fields to be between n_core_fields and n_fields. Do not support infimum,supremum. They are never supposed to be stored in dtuple_t, because page creation nowadays uses a lower-level method for initializing them. rec_convert_dtuple_to_rec_comp(): Assign the status bits based on the number of fields. btr_cur_trim(): In an update, trim the index entry as needed. For the 'default row', handle rollback specially. For user records, omit fields that match the 'default row'. btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete(): Skip locking and adaptive hash index for the 'default row'. row_log_table_apply_convert_mrec(): Replace 'default row' values if needed. In the temporary file that is applied by row_log_table_apply(), we must identify whether the records contain the extra header for instantly added columns. For now, we will allocate an additional byte for this for ROW_T_INSERT and ROW_T_UPDATE records when the source table has been subject to instant ADD COLUMN. The ROW_T_DELETE records are fine, as they will be converted and will only contain 'core' columns (PRIMARY KEY and some system columns) that are converted from dtuple_t. rec_get_converted_size_temp(), rec_init_offsets_temp(), rec_convert_dtuple_to_temp(): Add the parameter 'status'. REC_INFO_DEFAULT_ROW = REC_INFO_MIN_REC_FLAG | REC_STATUS_COLUMNS_ADDED: An info_bits constant for distinguishing the 'default row' record. rec_comp_status_t: An enum of the status bit values. rec_leaf_format: An enum that replaces the bool parameter of rec_init_offsets_comp_ordinary().
8 years ago
MDEV-21907: InnoDB: Enable -Wconversion on clang and GCC The -Wconversion in GCC seems to be stricter than in clang. GCC at least since version 4.4.7 issues truncation warnings for assignments to bitfields, while clang 10 appears to only issue warnings when the sizes in bytes rounded to the nearest integer powers of 2 are different. Before GCC 10.0.0, -Wconversion required more casts and would not allow some operations, such as x<<=1 or x+=1 on a data type that is narrower than int. GCC 5 (but not GCC 4, GCC 6, or any later version) is complaining about x|=y even when x and y are compatible types that are narrower than int. Hence, we must rewrite some x|=y as x=static_cast<byte>(x|y) or similar, or we must disable -Wconversion. In GCC 6 and later, the warning for assigning wider to bitfields that are narrower than 8, 16, or 32 bits can be suppressed by applying a bitwise & with the exact bitmask of the bitfield. For older GCC, we must disable -Wconversion for GCC 4 or 5 in such cases. The bitwise negation operator appears to promote short integers to a wider type, and hence we must add explicit truncation casts around them. Microsoft Visual C does not allow a static_cast to truncate a constant, such as static_cast<byte>(1) truncating int. Hence, we will use the constructor-style cast byte(~1) for such cases. This has been tested at least with GCC 4.8.5, 5.4.0, 7.4.0, 9.2.1, 10.0.0, clang 9.0.1, 10.0.0, and MSVC 14.22.27905 (Microsoft Visual Studio 2019) on 64-bit and 32-bit targets (IA-32, AMD64, POWER 8, POWER 9, ARMv8).
6 years ago
MDEV-21907: InnoDB: Enable -Wconversion on clang and GCC The -Wconversion in GCC seems to be stricter than in clang. GCC at least since version 4.4.7 issues truncation warnings for assignments to bitfields, while clang 10 appears to only issue warnings when the sizes in bytes rounded to the nearest integer powers of 2 are different. Before GCC 10.0.0, -Wconversion required more casts and would not allow some operations, such as x<<=1 or x+=1 on a data type that is narrower than int. GCC 5 (but not GCC 4, GCC 6, or any later version) is complaining about x|=y even when x and y are compatible types that are narrower than int. Hence, we must rewrite some x|=y as x=static_cast<byte>(x|y) or similar, or we must disable -Wconversion. In GCC 6 and later, the warning for assigning wider to bitfields that are narrower than 8, 16, or 32 bits can be suppressed by applying a bitwise & with the exact bitmask of the bitfield. For older GCC, we must disable -Wconversion for GCC 4 or 5 in such cases. The bitwise negation operator appears to promote short integers to a wider type, and hence we must add explicit truncation casts around them. Microsoft Visual C does not allow a static_cast to truncate a constant, such as static_cast<byte>(1) truncating int. Hence, we will use the constructor-style cast byte(~1) for such cases. This has been tested at least with GCC 4.8.5, 5.4.0, 7.4.0, 9.2.1, 10.0.0, clang 9.0.1, 10.0.0, and MSVC 14.22.27905 (Microsoft Visual Studio 2019) on 64-bit and 32-bit targets (IA-32, AMD64, POWER 8, POWER 9, ARMv8).
6 years ago
MDEV-21907: InnoDB: Enable -Wconversion on clang and GCC The -Wconversion in GCC seems to be stricter than in clang. GCC at least since version 4.4.7 issues truncation warnings for assignments to bitfields, while clang 10 appears to only issue warnings when the sizes in bytes rounded to the nearest integer powers of 2 are different. Before GCC 10.0.0, -Wconversion required more casts and would not allow some operations, such as x<<=1 or x+=1 on a data type that is narrower than int. GCC 5 (but not GCC 4, GCC 6, or any later version) is complaining about x|=y even when x and y are compatible types that are narrower than int. Hence, we must rewrite some x|=y as x=static_cast<byte>(x|y) or similar, or we must disable -Wconversion. In GCC 6 and later, the warning for assigning wider to bitfields that are narrower than 8, 16, or 32 bits can be suppressed by applying a bitwise & with the exact bitmask of the bitfield. For older GCC, we must disable -Wconversion for GCC 4 or 5 in such cases. The bitwise negation operator appears to promote short integers to a wider type, and hence we must add explicit truncation casts around them. Microsoft Visual C does not allow a static_cast to truncate a constant, such as static_cast<byte>(1) truncating int. Hence, we will use the constructor-style cast byte(~1) for such cases. This has been tested at least with GCC 4.8.5, 5.4.0, 7.4.0, 9.2.1, 10.0.0, clang 9.0.1, 10.0.0, and MSVC 14.22.27905 (Microsoft Visual Studio 2019) on 64-bit and 32-bit targets (IA-32, AMD64, POWER 8, POWER 9, ARMv8).
6 years ago
MDEV-24971 InnoDB access freed virtual column after rollback of secondary index Problem: ======== InnoDB fails to clean the index stub if it fails to add the virtual index which contains new virtual column. But it clears the newly virtual column from index in clear_added_indexes() during inplace_alter_table. On commit, InnoDB evicts and reload the table. In case of rollback, it doesn't happen. InnoDB clears the ABORTED index while opening the table or doing the DDL. In the mean time, InnoDB can access the dropped virtual index columns while creating prebuilt or rollback of concurrent DML. Solution: ========== (1) InnoDB should maintain newly added virtual column while rollbacking the newly added virtual index. (2) InnoDB must not defer the index removal if the alter table is executed with LOCK=EXCLUSIVE. (3) For LOCK=SHARED, InnoDB should check whether the table has any other transaction lock other than alter transaction before deferring the index stub. Replaced has_new_v_col with dict_add_vcol_info in dict_index_t to indicate whether the index has any new virtual column. dict_index_t::has_new_v_col(): Returns whether the index has newly added virtual column, it doesn't say which columns are newly added virtual column ha_innobase_inplace_ctx::is_new_vcol(): Return whether the given column is added as a part of the current alter. ha_innobase_inplace_ctx::clean_new_vcol_index(): Copy the newly added virtual column to new_vcol_info in dict_index_t. Replace the column in the index fields with virtual column stored in new_vcol_info. dict_index_t::assign_new_v_col(): Store the number of virtual column added in index as a part of alter table. dict_index_t::get_n_new_vcol(): Get the number of newly added virtual column dict_index_t::assign_drop_v_col(): Allocate the memory for adding new virtual column in new_vcol_info. dict_index_t::add_drop_v_col(): Add the newly added virtual column in new_vcol_info. dict_table_t::has_lock_for_other_trx(): Whether the table has any other transaction lock than given transaction. row_merge_drop_indexes(): Add parameter alter_trx and check whether the table has any other lock than alter transaction.
5 years ago
MDEV-11369 Instant ADD COLUMN for InnoDB For InnoDB tables, adding, dropping and reordering columns has required a rebuild of the table and all its indexes. Since MySQL 5.6 (and MariaDB 10.0) this has been supported online (LOCK=NONE), allowing concurrent modification of the tables. This work revises the InnoDB ROW_FORMAT=REDUNDANT, ROW_FORMAT=COMPACT and ROW_FORMAT=DYNAMIC so that columns can be appended instantaneously, with only minor changes performed to the table structure. The counter innodb_instant_alter_column in INFORMATION_SCHEMA.GLOBAL_STATUS is incremented whenever a table rebuild operation is converted into an instant ADD COLUMN operation. ROW_FORMAT=COMPRESSED tables will not support instant ADD COLUMN. Some usability limitations will be addressed in subsequent work: MDEV-13134 Introduce ALTER TABLE attributes ALGORITHM=NOCOPY and ALGORITHM=INSTANT MDEV-14016 Allow instant ADD COLUMN, ADD INDEX, LOCK=NONE The format of the clustered index (PRIMARY KEY) is changed as follows: (1) The FIL_PAGE_TYPE of the root page will be FIL_PAGE_TYPE_INSTANT, and a new field PAGE_INSTANT will contain the original number of fields in the clustered index ('core' fields). If instant ADD COLUMN has not been used or the table becomes empty, or the very first instant ADD COLUMN operation is rolled back, the fields PAGE_INSTANT and FIL_PAGE_TYPE will be reset to 0 and FIL_PAGE_INDEX. (2) A special 'default row' record is inserted into the leftmost leaf, between the page infimum and the first user record. This record is distinguished by the REC_INFO_MIN_REC_FLAG, and it is otherwise in the same format as records that contain values for the instantly added columns. This 'default row' always has the same number of fields as the clustered index according to the table definition. The values of 'core' fields are to be ignored. For other fields, the 'default row' will contain the default values as they were during the ALTER TABLE statement. (If the column default values are changed later, those values will only be stored in the .frm file. The 'default row' will contain the original evaluated values, which must be the same for every row.) The 'default row' must be completely hidden from higher-level access routines. Assertions have been added to ensure that no 'default row' is ever present in the adaptive hash index or in locked records. The 'default row' is never delete-marked. (3) In clustered index leaf page records, the number of fields must reside between the number of 'core' fields (dict_index_t::n_core_fields introduced in this work) and dict_index_t::n_fields. If the number of fields is less than dict_index_t::n_fields, the missing fields are replaced with the column value of the 'default row'. Note: The number of fields in the record may shrink if some of the last instantly added columns are updated to the value that is in the 'default row'. The function btr_cur_trim() implements this 'compression' on update and rollback; dtuple::trim() implements it on insert. (4) In ROW_FORMAT=COMPACT and ROW_FORMAT=DYNAMIC records, the new status value REC_STATUS_COLUMNS_ADDED will indicate the presence of a new record header that will encode n_fields-n_core_fields-1 in 1 or 2 bytes. (In ROW_FORMAT=REDUNDANT records, the record header always explicitly encodes the number of fields.) We introduce the undo log record type TRX_UNDO_INSERT_DEFAULT for covering the insert of the 'default row' record when instant ADD COLUMN is used for the first time. Subsequent instant ADD COLUMN can use TRX_UNDO_UPD_EXIST_REC. This is joint work with Vin Chen (陈福荣) from Tencent. The design that was discussed in April 2017 would not have allowed import or export of data files, because instead of the 'default row' it would have introduced a data dictionary table. The test rpl.rpl_alter_instant is exactly as contributed in pull request #408. The test innodb.instant_alter is based on a contributed test. The redo log record format changes for ROW_FORMAT=DYNAMIC and ROW_FORMAT=COMPACT are as contributed. (With this change present, crash recovery from MariaDB 10.3.1 will fail in spectacular ways!) Also the semantics of higher-level redo log records that modify the PAGE_INSTANT field is changed. The redo log format version identifier was already changed to LOG_HEADER_FORMAT_CURRENT=103 in MariaDB 10.3.1. Everything else has been rewritten by me. Thanks to Elena Stepanova, the code has been tested extensively. When rolling back an instant ADD COLUMN operation, we must empty the PAGE_FREE list after deleting or shortening the 'default row' record, by calling either btr_page_empty() or btr_page_reorganize(). We must know the size of each entry in the PAGE_FREE list. If rollback left a freed copy of the 'default row' in the PAGE_FREE list, we would be unable to determine its size (if it is in ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC) because it would contain more fields than the rolled-back definition of the clustered index. UNIV_SQL_DEFAULT: A new special constant that designates an instantly added column that is not present in the clustered index record. len_is_stored(): Check if a length is an actual length. There are two magic length values: UNIV_SQL_DEFAULT, UNIV_SQL_NULL. dict_col_t::def_val: The 'default row' value of the column. If the column is not added instantly, def_val.len will be UNIV_SQL_DEFAULT. dict_col_t: Add the accessors is_virtual(), is_nullable(), is_instant(), instant_value(). dict_col_t::remove_instant(): Remove the 'instant ADD' status of a column. dict_col_t::name(const dict_table_t& table): Replaces dict_table_get_col_name(). dict_index_t::n_core_fields: The original number of fields. For secondary indexes and if instant ADD COLUMN has not been used, this will be equal to dict_index_t::n_fields. dict_index_t::n_core_null_bytes: Number of bytes needed to represent the null flags; usually equal to UT_BITS_IN_BYTES(n_nullable). dict_index_t::NO_CORE_NULL_BYTES: Magic value signalling that n_core_null_bytes was not initialized yet from the clustered index root page. dict_index_t: Add the accessors is_instant(), is_clust(), get_n_nullable(), instant_field_value(). dict_index_t::instant_add_field(): Adjust clustered index metadata for instant ADD COLUMN. dict_index_t::remove_instant(): Remove the 'instant ADD' status of a clustered index when the table becomes empty, or the very first instant ADD COLUMN operation is rolled back. dict_table_t: Add the accessors is_instant(), is_temporary(), supports_instant(). dict_table_t::instant_add_column(): Adjust metadata for instant ADD COLUMN. dict_table_t::rollback_instant(): Adjust metadata on the rollback of instant ADD COLUMN. prepare_inplace_alter_table_dict(): First create the ctx->new_table, and only then decide if the table really needs to be rebuilt. We must split the creation of table or index metadata from the creation of the dictionary table records and the creation of the data. In this way, we can transform a table-rebuilding operation into an instant ADD COLUMN operation. Dictionary objects will only be added to cache when table rebuilding or index creation is needed. The ctx->instant_table will never be added to cache. dict_table_t::add_to_cache(): Modified and renamed from dict_table_add_to_cache(). Do not modify the table metadata. Let the callers invoke dict_table_add_system_columns() and if needed, set can_be_evicted. dict_create_sys_tables_tuple(), dict_create_table_step(): Omit the system columns (which will now exist in the dict_table_t object already at this point). dict_create_table_step(): Expect the callers to invoke dict_table_add_system_columns(). pars_create_table(): Before creating the table creation execution graph, invoke dict_table_add_system_columns(). row_create_table_for_mysql(): Expect all callers to invoke dict_table_add_system_columns(). create_index_dict(): Replaces row_merge_create_index_graph(). innodb_update_n_cols(): Renamed from innobase_update_n_virtual(). Call my_error() if an error occurs. btr_cur_instant_init(), btr_cur_instant_init_low(), btr_cur_instant_root_init(): Load additional metadata from the clustered index and set dict_index_t::n_core_null_bytes. This is invoked when table metadata is first loaded into the data dictionary. dict_boot(): Initialize n_core_null_bytes for the four hard-coded dictionary tables. dict_create_index_step(): Initialize n_core_null_bytes. This is executed as part of CREATE TABLE. dict_index_build_internal_clust(): Initialize n_core_null_bytes to NO_CORE_NULL_BYTES if table->supports_instant(). row_create_index_for_mysql(): Initialize n_core_null_bytes for CREATE TEMPORARY TABLE. commit_cache_norebuild(): Call the code to rename or enlarge columns in the cache only if instant ADD COLUMN is not being used. (Instant ADD COLUMN would copy all column metadata from instant_table to old_table, including the names and lengths.) PAGE_INSTANT: A new 13-bit field for storing dict_index_t::n_core_fields. This is repurposing the 16-bit field PAGE_DIRECTION, of which only the least significant 3 bits were used. The original byte containing PAGE_DIRECTION will be accessible via the new constant PAGE_DIRECTION_B. page_get_instant(), page_set_instant(): Accessors for the PAGE_INSTANT. page_ptr_get_direction(), page_get_direction(), page_ptr_set_direction(): Accessors for PAGE_DIRECTION. page_direction_reset(): Reset PAGE_DIRECTION, PAGE_N_DIRECTION. page_direction_increment(): Increment PAGE_N_DIRECTION and set PAGE_DIRECTION. rec_get_offsets(): Use the 'leaf' parameter for non-debug purposes, and assume that heap_no is always set. Initialize all dict_index_t::n_fields for ROW_FORMAT=REDUNDANT records, even if the record contains fewer fields. rec_offs_make_valid(): Add the parameter 'leaf'. rec_copy_prefix_to_dtuple(): Assert that the tuple is only built on the core fields. Instant ADD COLUMN only applies to the clustered index, and we should never build a search key that has more than the PRIMARY KEY and possibly DB_TRX_ID,DB_ROLL_PTR. All these columns are always present. dict_index_build_data_tuple(): Remove assertions that would be duplicated in rec_copy_prefix_to_dtuple(). rec_init_offsets(): Support ROW_FORMAT=REDUNDANT records whose number of fields is between n_core_fields and n_fields. cmp_rec_rec_with_match(): Implement the comparison between two MIN_REC_FLAG records. trx_t::in_rollback: Make the field available in non-debug builds. trx_start_for_ddl_low(): Remove dangerous error-tolerance. A dictionary transaction must be flagged as such before it has generated any undo log records. This is because trx_undo_assign_undo() will mark the transaction as a dictionary transaction in the undo log header right before the very first undo log record is being written. btr_index_rec_validate(): Account for instant ADD COLUMN row_undo_ins_remove_clust_rec(): On the rollback of an insert into SYS_COLUMNS, revert instant ADD COLUMN in the cache by removing the last column from the table and the clustered index. row_search_on_row_ref(), row_undo_mod_parse_undo_rec(), row_undo_mod(), trx_undo_update_rec_get_update(): Handle the 'default row' as a special case. dtuple_t::trim(index): Omit a redundant suffix of an index tuple right before insert or update. After instant ADD COLUMN, if the last fields of a clustered index tuple match the 'default row', there is no need to store them. While trimming the entry, we must hold a page latch, so that the table cannot be emptied and the 'default row' be deleted. btr_cur_optimistic_update(), btr_cur_pessimistic_update(), row_upd_clust_rec_by_insert(), row_ins_clust_index_entry_low(): Invoke dtuple_t::trim() if needed. row_ins_clust_index_entry(): Restore dtuple_t::n_fields after calling row_ins_clust_index_entry_low(). rec_get_converted_size(), rec_get_converted_size_comp(): Allow the number of fields to be between n_core_fields and n_fields. Do not support infimum,supremum. They are never supposed to be stored in dtuple_t, because page creation nowadays uses a lower-level method for initializing them. rec_convert_dtuple_to_rec_comp(): Assign the status bits based on the number of fields. btr_cur_trim(): In an update, trim the index entry as needed. For the 'default row', handle rollback specially. For user records, omit fields that match the 'default row'. btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete(): Skip locking and adaptive hash index for the 'default row'. row_log_table_apply_convert_mrec(): Replace 'default row' values if needed. In the temporary file that is applied by row_log_table_apply(), we must identify whether the records contain the extra header for instantly added columns. For now, we will allocate an additional byte for this for ROW_T_INSERT and ROW_T_UPDATE records when the source table has been subject to instant ADD COLUMN. The ROW_T_DELETE records are fine, as they will be converted and will only contain 'core' columns (PRIMARY KEY and some system columns) that are converted from dtuple_t. rec_get_converted_size_temp(), rec_init_offsets_temp(), rec_convert_dtuple_to_temp(): Add the parameter 'status'. REC_INFO_DEFAULT_ROW = REC_INFO_MIN_REC_FLAG | REC_STATUS_COLUMNS_ADDED: An info_bits constant for distinguishing the 'default row' record. rec_comp_status_t: An enum of the status bit values. rec_leaf_format: An enum that replaces the bool parameter of rec_init_offsets_comp_ordinary().
8 years ago
MDEV-11369 Instant ADD COLUMN for InnoDB For InnoDB tables, adding, dropping and reordering columns has required a rebuild of the table and all its indexes. Since MySQL 5.6 (and MariaDB 10.0) this has been supported online (LOCK=NONE), allowing concurrent modification of the tables. This work revises the InnoDB ROW_FORMAT=REDUNDANT, ROW_FORMAT=COMPACT and ROW_FORMAT=DYNAMIC so that columns can be appended instantaneously, with only minor changes performed to the table structure. The counter innodb_instant_alter_column in INFORMATION_SCHEMA.GLOBAL_STATUS is incremented whenever a table rebuild operation is converted into an instant ADD COLUMN operation. ROW_FORMAT=COMPRESSED tables will not support instant ADD COLUMN. Some usability limitations will be addressed in subsequent work: MDEV-13134 Introduce ALTER TABLE attributes ALGORITHM=NOCOPY and ALGORITHM=INSTANT MDEV-14016 Allow instant ADD COLUMN, ADD INDEX, LOCK=NONE The format of the clustered index (PRIMARY KEY) is changed as follows: (1) The FIL_PAGE_TYPE of the root page will be FIL_PAGE_TYPE_INSTANT, and a new field PAGE_INSTANT will contain the original number of fields in the clustered index ('core' fields). If instant ADD COLUMN has not been used or the table becomes empty, or the very first instant ADD COLUMN operation is rolled back, the fields PAGE_INSTANT and FIL_PAGE_TYPE will be reset to 0 and FIL_PAGE_INDEX. (2) A special 'default row' record is inserted into the leftmost leaf, between the page infimum and the first user record. This record is distinguished by the REC_INFO_MIN_REC_FLAG, and it is otherwise in the same format as records that contain values for the instantly added columns. This 'default row' always has the same number of fields as the clustered index according to the table definition. The values of 'core' fields are to be ignored. For other fields, the 'default row' will contain the default values as they were during the ALTER TABLE statement. (If the column default values are changed later, those values will only be stored in the .frm file. The 'default row' will contain the original evaluated values, which must be the same for every row.) The 'default row' must be completely hidden from higher-level access routines. Assertions have been added to ensure that no 'default row' is ever present in the adaptive hash index or in locked records. The 'default row' is never delete-marked. (3) In clustered index leaf page records, the number of fields must reside between the number of 'core' fields (dict_index_t::n_core_fields introduced in this work) and dict_index_t::n_fields. If the number of fields is less than dict_index_t::n_fields, the missing fields are replaced with the column value of the 'default row'. Note: The number of fields in the record may shrink if some of the last instantly added columns are updated to the value that is in the 'default row'. The function btr_cur_trim() implements this 'compression' on update and rollback; dtuple::trim() implements it on insert. (4) In ROW_FORMAT=COMPACT and ROW_FORMAT=DYNAMIC records, the new status value REC_STATUS_COLUMNS_ADDED will indicate the presence of a new record header that will encode n_fields-n_core_fields-1 in 1 or 2 bytes. (In ROW_FORMAT=REDUNDANT records, the record header always explicitly encodes the number of fields.) We introduce the undo log record type TRX_UNDO_INSERT_DEFAULT for covering the insert of the 'default row' record when instant ADD COLUMN is used for the first time. Subsequent instant ADD COLUMN can use TRX_UNDO_UPD_EXIST_REC. This is joint work with Vin Chen (陈福荣) from Tencent. The design that was discussed in April 2017 would not have allowed import or export of data files, because instead of the 'default row' it would have introduced a data dictionary table. The test rpl.rpl_alter_instant is exactly as contributed in pull request #408. The test innodb.instant_alter is based on a contributed test. The redo log record format changes for ROW_FORMAT=DYNAMIC and ROW_FORMAT=COMPACT are as contributed. (With this change present, crash recovery from MariaDB 10.3.1 will fail in spectacular ways!) Also the semantics of higher-level redo log records that modify the PAGE_INSTANT field is changed. The redo log format version identifier was already changed to LOG_HEADER_FORMAT_CURRENT=103 in MariaDB 10.3.1. Everything else has been rewritten by me. Thanks to Elena Stepanova, the code has been tested extensively. When rolling back an instant ADD COLUMN operation, we must empty the PAGE_FREE list after deleting or shortening the 'default row' record, by calling either btr_page_empty() or btr_page_reorganize(). We must know the size of each entry in the PAGE_FREE list. If rollback left a freed copy of the 'default row' in the PAGE_FREE list, we would be unable to determine its size (if it is in ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC) because it would contain more fields than the rolled-back definition of the clustered index. UNIV_SQL_DEFAULT: A new special constant that designates an instantly added column that is not present in the clustered index record. len_is_stored(): Check if a length is an actual length. There are two magic length values: UNIV_SQL_DEFAULT, UNIV_SQL_NULL. dict_col_t::def_val: The 'default row' value of the column. If the column is not added instantly, def_val.len will be UNIV_SQL_DEFAULT. dict_col_t: Add the accessors is_virtual(), is_nullable(), is_instant(), instant_value(). dict_col_t::remove_instant(): Remove the 'instant ADD' status of a column. dict_col_t::name(const dict_table_t& table): Replaces dict_table_get_col_name(). dict_index_t::n_core_fields: The original number of fields. For secondary indexes and if instant ADD COLUMN has not been used, this will be equal to dict_index_t::n_fields. dict_index_t::n_core_null_bytes: Number of bytes needed to represent the null flags; usually equal to UT_BITS_IN_BYTES(n_nullable). dict_index_t::NO_CORE_NULL_BYTES: Magic value signalling that n_core_null_bytes was not initialized yet from the clustered index root page. dict_index_t: Add the accessors is_instant(), is_clust(), get_n_nullable(), instant_field_value(). dict_index_t::instant_add_field(): Adjust clustered index metadata for instant ADD COLUMN. dict_index_t::remove_instant(): Remove the 'instant ADD' status of a clustered index when the table becomes empty, or the very first instant ADD COLUMN operation is rolled back. dict_table_t: Add the accessors is_instant(), is_temporary(), supports_instant(). dict_table_t::instant_add_column(): Adjust metadata for instant ADD COLUMN. dict_table_t::rollback_instant(): Adjust metadata on the rollback of instant ADD COLUMN. prepare_inplace_alter_table_dict(): First create the ctx->new_table, and only then decide if the table really needs to be rebuilt. We must split the creation of table or index metadata from the creation of the dictionary table records and the creation of the data. In this way, we can transform a table-rebuilding operation into an instant ADD COLUMN operation. Dictionary objects will only be added to cache when table rebuilding or index creation is needed. The ctx->instant_table will never be added to cache. dict_table_t::add_to_cache(): Modified and renamed from dict_table_add_to_cache(). Do not modify the table metadata. Let the callers invoke dict_table_add_system_columns() and if needed, set can_be_evicted. dict_create_sys_tables_tuple(), dict_create_table_step(): Omit the system columns (which will now exist in the dict_table_t object already at this point). dict_create_table_step(): Expect the callers to invoke dict_table_add_system_columns(). pars_create_table(): Before creating the table creation execution graph, invoke dict_table_add_system_columns(). row_create_table_for_mysql(): Expect all callers to invoke dict_table_add_system_columns(). create_index_dict(): Replaces row_merge_create_index_graph(). innodb_update_n_cols(): Renamed from innobase_update_n_virtual(). Call my_error() if an error occurs. btr_cur_instant_init(), btr_cur_instant_init_low(), btr_cur_instant_root_init(): Load additional metadata from the clustered index and set dict_index_t::n_core_null_bytes. This is invoked when table metadata is first loaded into the data dictionary. dict_boot(): Initialize n_core_null_bytes for the four hard-coded dictionary tables. dict_create_index_step(): Initialize n_core_null_bytes. This is executed as part of CREATE TABLE. dict_index_build_internal_clust(): Initialize n_core_null_bytes to NO_CORE_NULL_BYTES if table->supports_instant(). row_create_index_for_mysql(): Initialize n_core_null_bytes for CREATE TEMPORARY TABLE. commit_cache_norebuild(): Call the code to rename or enlarge columns in the cache only if instant ADD COLUMN is not being used. (Instant ADD COLUMN would copy all column metadata from instant_table to old_table, including the names and lengths.) PAGE_INSTANT: A new 13-bit field for storing dict_index_t::n_core_fields. This is repurposing the 16-bit field PAGE_DIRECTION, of which only the least significant 3 bits were used. The original byte containing PAGE_DIRECTION will be accessible via the new constant PAGE_DIRECTION_B. page_get_instant(), page_set_instant(): Accessors for the PAGE_INSTANT. page_ptr_get_direction(), page_get_direction(), page_ptr_set_direction(): Accessors for PAGE_DIRECTION. page_direction_reset(): Reset PAGE_DIRECTION, PAGE_N_DIRECTION. page_direction_increment(): Increment PAGE_N_DIRECTION and set PAGE_DIRECTION. rec_get_offsets(): Use the 'leaf' parameter for non-debug purposes, and assume that heap_no is always set. Initialize all dict_index_t::n_fields for ROW_FORMAT=REDUNDANT records, even if the record contains fewer fields. rec_offs_make_valid(): Add the parameter 'leaf'. rec_copy_prefix_to_dtuple(): Assert that the tuple is only built on the core fields. Instant ADD COLUMN only applies to the clustered index, and we should never build a search key that has more than the PRIMARY KEY and possibly DB_TRX_ID,DB_ROLL_PTR. All these columns are always present. dict_index_build_data_tuple(): Remove assertions that would be duplicated in rec_copy_prefix_to_dtuple(). rec_init_offsets(): Support ROW_FORMAT=REDUNDANT records whose number of fields is between n_core_fields and n_fields. cmp_rec_rec_with_match(): Implement the comparison between two MIN_REC_FLAG records. trx_t::in_rollback: Make the field available in non-debug builds. trx_start_for_ddl_low(): Remove dangerous error-tolerance. A dictionary transaction must be flagged as such before it has generated any undo log records. This is because trx_undo_assign_undo() will mark the transaction as a dictionary transaction in the undo log header right before the very first undo log record is being written. btr_index_rec_validate(): Account for instant ADD COLUMN row_undo_ins_remove_clust_rec(): On the rollback of an insert into SYS_COLUMNS, revert instant ADD COLUMN in the cache by removing the last column from the table and the clustered index. row_search_on_row_ref(), row_undo_mod_parse_undo_rec(), row_undo_mod(), trx_undo_update_rec_get_update(): Handle the 'default row' as a special case. dtuple_t::trim(index): Omit a redundant suffix of an index tuple right before insert or update. After instant ADD COLUMN, if the last fields of a clustered index tuple match the 'default row', there is no need to store them. While trimming the entry, we must hold a page latch, so that the table cannot be emptied and the 'default row' be deleted. btr_cur_optimistic_update(), btr_cur_pessimistic_update(), row_upd_clust_rec_by_insert(), row_ins_clust_index_entry_low(): Invoke dtuple_t::trim() if needed. row_ins_clust_index_entry(): Restore dtuple_t::n_fields after calling row_ins_clust_index_entry_low(). rec_get_converted_size(), rec_get_converted_size_comp(): Allow the number of fields to be between n_core_fields and n_fields. Do not support infimum,supremum. They are never supposed to be stored in dtuple_t, because page creation nowadays uses a lower-level method for initializing them. rec_convert_dtuple_to_rec_comp(): Assign the status bits based on the number of fields. btr_cur_trim(): In an update, trim the index entry as needed. For the 'default row', handle rollback specially. For user records, omit fields that match the 'default row'. btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete(): Skip locking and adaptive hash index for the 'default row'. row_log_table_apply_convert_mrec(): Replace 'default row' values if needed. In the temporary file that is applied by row_log_table_apply(), we must identify whether the records contain the extra header for instantly added columns. For now, we will allocate an additional byte for this for ROW_T_INSERT and ROW_T_UPDATE records when the source table has been subject to instant ADD COLUMN. The ROW_T_DELETE records are fine, as they will be converted and will only contain 'core' columns (PRIMARY KEY and some system columns) that are converted from dtuple_t. rec_get_converted_size_temp(), rec_init_offsets_temp(), rec_convert_dtuple_to_temp(): Add the parameter 'status'. REC_INFO_DEFAULT_ROW = REC_INFO_MIN_REC_FLAG | REC_STATUS_COLUMNS_ADDED: An info_bits constant for distinguishing the 'default row' record. rec_comp_status_t: An enum of the status bit values. rec_leaf_format: An enum that replaces the bool parameter of rec_init_offsets_comp_ordinary().
8 years ago
MDEV-11369 Instant ADD COLUMN for InnoDB For InnoDB tables, adding, dropping and reordering columns has required a rebuild of the table and all its indexes. Since MySQL 5.6 (and MariaDB 10.0) this has been supported online (LOCK=NONE), allowing concurrent modification of the tables. This work revises the InnoDB ROW_FORMAT=REDUNDANT, ROW_FORMAT=COMPACT and ROW_FORMAT=DYNAMIC so that columns can be appended instantaneously, with only minor changes performed to the table structure. The counter innodb_instant_alter_column in INFORMATION_SCHEMA.GLOBAL_STATUS is incremented whenever a table rebuild operation is converted into an instant ADD COLUMN operation. ROW_FORMAT=COMPRESSED tables will not support instant ADD COLUMN. Some usability limitations will be addressed in subsequent work: MDEV-13134 Introduce ALTER TABLE attributes ALGORITHM=NOCOPY and ALGORITHM=INSTANT MDEV-14016 Allow instant ADD COLUMN, ADD INDEX, LOCK=NONE The format of the clustered index (PRIMARY KEY) is changed as follows: (1) The FIL_PAGE_TYPE of the root page will be FIL_PAGE_TYPE_INSTANT, and a new field PAGE_INSTANT will contain the original number of fields in the clustered index ('core' fields). If instant ADD COLUMN has not been used or the table becomes empty, or the very first instant ADD COLUMN operation is rolled back, the fields PAGE_INSTANT and FIL_PAGE_TYPE will be reset to 0 and FIL_PAGE_INDEX. (2) A special 'default row' record is inserted into the leftmost leaf, between the page infimum and the first user record. This record is distinguished by the REC_INFO_MIN_REC_FLAG, and it is otherwise in the same format as records that contain values for the instantly added columns. This 'default row' always has the same number of fields as the clustered index according to the table definition. The values of 'core' fields are to be ignored. For other fields, the 'default row' will contain the default values as they were during the ALTER TABLE statement. (If the column default values are changed later, those values will only be stored in the .frm file. The 'default row' will contain the original evaluated values, which must be the same for every row.) The 'default row' must be completely hidden from higher-level access routines. Assertions have been added to ensure that no 'default row' is ever present in the adaptive hash index or in locked records. The 'default row' is never delete-marked. (3) In clustered index leaf page records, the number of fields must reside between the number of 'core' fields (dict_index_t::n_core_fields introduced in this work) and dict_index_t::n_fields. If the number of fields is less than dict_index_t::n_fields, the missing fields are replaced with the column value of the 'default row'. Note: The number of fields in the record may shrink if some of the last instantly added columns are updated to the value that is in the 'default row'. The function btr_cur_trim() implements this 'compression' on update and rollback; dtuple::trim() implements it on insert. (4) In ROW_FORMAT=COMPACT and ROW_FORMAT=DYNAMIC records, the new status value REC_STATUS_COLUMNS_ADDED will indicate the presence of a new record header that will encode n_fields-n_core_fields-1 in 1 or 2 bytes. (In ROW_FORMAT=REDUNDANT records, the record header always explicitly encodes the number of fields.) We introduce the undo log record type TRX_UNDO_INSERT_DEFAULT for covering the insert of the 'default row' record when instant ADD COLUMN is used for the first time. Subsequent instant ADD COLUMN can use TRX_UNDO_UPD_EXIST_REC. This is joint work with Vin Chen (陈福荣) from Tencent. The design that was discussed in April 2017 would not have allowed import or export of data files, because instead of the 'default row' it would have introduced a data dictionary table. The test rpl.rpl_alter_instant is exactly as contributed in pull request #408. The test innodb.instant_alter is based on a contributed test. The redo log record format changes for ROW_FORMAT=DYNAMIC and ROW_FORMAT=COMPACT are as contributed. (With this change present, crash recovery from MariaDB 10.3.1 will fail in spectacular ways!) Also the semantics of higher-level redo log records that modify the PAGE_INSTANT field is changed. The redo log format version identifier was already changed to LOG_HEADER_FORMAT_CURRENT=103 in MariaDB 10.3.1. Everything else has been rewritten by me. Thanks to Elena Stepanova, the code has been tested extensively. When rolling back an instant ADD COLUMN operation, we must empty the PAGE_FREE list after deleting or shortening the 'default row' record, by calling either btr_page_empty() or btr_page_reorganize(). We must know the size of each entry in the PAGE_FREE list. If rollback left a freed copy of the 'default row' in the PAGE_FREE list, we would be unable to determine its size (if it is in ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC) because it would contain more fields than the rolled-back definition of the clustered index. UNIV_SQL_DEFAULT: A new special constant that designates an instantly added column that is not present in the clustered index record. len_is_stored(): Check if a length is an actual length. There are two magic length values: UNIV_SQL_DEFAULT, UNIV_SQL_NULL. dict_col_t::def_val: The 'default row' value of the column. If the column is not added instantly, def_val.len will be UNIV_SQL_DEFAULT. dict_col_t: Add the accessors is_virtual(), is_nullable(), is_instant(), instant_value(). dict_col_t::remove_instant(): Remove the 'instant ADD' status of a column. dict_col_t::name(const dict_table_t& table): Replaces dict_table_get_col_name(). dict_index_t::n_core_fields: The original number of fields. For secondary indexes and if instant ADD COLUMN has not been used, this will be equal to dict_index_t::n_fields. dict_index_t::n_core_null_bytes: Number of bytes needed to represent the null flags; usually equal to UT_BITS_IN_BYTES(n_nullable). dict_index_t::NO_CORE_NULL_BYTES: Magic value signalling that n_core_null_bytes was not initialized yet from the clustered index root page. dict_index_t: Add the accessors is_instant(), is_clust(), get_n_nullable(), instant_field_value(). dict_index_t::instant_add_field(): Adjust clustered index metadata for instant ADD COLUMN. dict_index_t::remove_instant(): Remove the 'instant ADD' status of a clustered index when the table becomes empty, or the very first instant ADD COLUMN operation is rolled back. dict_table_t: Add the accessors is_instant(), is_temporary(), supports_instant(). dict_table_t::instant_add_column(): Adjust metadata for instant ADD COLUMN. dict_table_t::rollback_instant(): Adjust metadata on the rollback of instant ADD COLUMN. prepare_inplace_alter_table_dict(): First create the ctx->new_table, and only then decide if the table really needs to be rebuilt. We must split the creation of table or index metadata from the creation of the dictionary table records and the creation of the data. In this way, we can transform a table-rebuilding operation into an instant ADD COLUMN operation. Dictionary objects will only be added to cache when table rebuilding or index creation is needed. The ctx->instant_table will never be added to cache. dict_table_t::add_to_cache(): Modified and renamed from dict_table_add_to_cache(). Do not modify the table metadata. Let the callers invoke dict_table_add_system_columns() and if needed, set can_be_evicted. dict_create_sys_tables_tuple(), dict_create_table_step(): Omit the system columns (which will now exist in the dict_table_t object already at this point). dict_create_table_step(): Expect the callers to invoke dict_table_add_system_columns(). pars_create_table(): Before creating the table creation execution graph, invoke dict_table_add_system_columns(). row_create_table_for_mysql(): Expect all callers to invoke dict_table_add_system_columns(). create_index_dict(): Replaces row_merge_create_index_graph(). innodb_update_n_cols(): Renamed from innobase_update_n_virtual(). Call my_error() if an error occurs. btr_cur_instant_init(), btr_cur_instant_init_low(), btr_cur_instant_root_init(): Load additional metadata from the clustered index and set dict_index_t::n_core_null_bytes. This is invoked when table metadata is first loaded into the data dictionary. dict_boot(): Initialize n_core_null_bytes for the four hard-coded dictionary tables. dict_create_index_step(): Initialize n_core_null_bytes. This is executed as part of CREATE TABLE. dict_index_build_internal_clust(): Initialize n_core_null_bytes to NO_CORE_NULL_BYTES if table->supports_instant(). row_create_index_for_mysql(): Initialize n_core_null_bytes for CREATE TEMPORARY TABLE. commit_cache_norebuild(): Call the code to rename or enlarge columns in the cache only if instant ADD COLUMN is not being used. (Instant ADD COLUMN would copy all column metadata from instant_table to old_table, including the names and lengths.) PAGE_INSTANT: A new 13-bit field for storing dict_index_t::n_core_fields. This is repurposing the 16-bit field PAGE_DIRECTION, of which only the least significant 3 bits were used. The original byte containing PAGE_DIRECTION will be accessible via the new constant PAGE_DIRECTION_B. page_get_instant(), page_set_instant(): Accessors for the PAGE_INSTANT. page_ptr_get_direction(), page_get_direction(), page_ptr_set_direction(): Accessors for PAGE_DIRECTION. page_direction_reset(): Reset PAGE_DIRECTION, PAGE_N_DIRECTION. page_direction_increment(): Increment PAGE_N_DIRECTION and set PAGE_DIRECTION. rec_get_offsets(): Use the 'leaf' parameter for non-debug purposes, and assume that heap_no is always set. Initialize all dict_index_t::n_fields for ROW_FORMAT=REDUNDANT records, even if the record contains fewer fields. rec_offs_make_valid(): Add the parameter 'leaf'. rec_copy_prefix_to_dtuple(): Assert that the tuple is only built on the core fields. Instant ADD COLUMN only applies to the clustered index, and we should never build a search key that has more than the PRIMARY KEY and possibly DB_TRX_ID,DB_ROLL_PTR. All these columns are always present. dict_index_build_data_tuple(): Remove assertions that would be duplicated in rec_copy_prefix_to_dtuple(). rec_init_offsets(): Support ROW_FORMAT=REDUNDANT records whose number of fields is between n_core_fields and n_fields. cmp_rec_rec_with_match(): Implement the comparison between two MIN_REC_FLAG records. trx_t::in_rollback: Make the field available in non-debug builds. trx_start_for_ddl_low(): Remove dangerous error-tolerance. A dictionary transaction must be flagged as such before it has generated any undo log records. This is because trx_undo_assign_undo() will mark the transaction as a dictionary transaction in the undo log header right before the very first undo log record is being written. btr_index_rec_validate(): Account for instant ADD COLUMN row_undo_ins_remove_clust_rec(): On the rollback of an insert into SYS_COLUMNS, revert instant ADD COLUMN in the cache by removing the last column from the table and the clustered index. row_search_on_row_ref(), row_undo_mod_parse_undo_rec(), row_undo_mod(), trx_undo_update_rec_get_update(): Handle the 'default row' as a special case. dtuple_t::trim(index): Omit a redundant suffix of an index tuple right before insert or update. After instant ADD COLUMN, if the last fields of a clustered index tuple match the 'default row', there is no need to store them. While trimming the entry, we must hold a page latch, so that the table cannot be emptied and the 'default row' be deleted. btr_cur_optimistic_update(), btr_cur_pessimistic_update(), row_upd_clust_rec_by_insert(), row_ins_clust_index_entry_low(): Invoke dtuple_t::trim() if needed. row_ins_clust_index_entry(): Restore dtuple_t::n_fields after calling row_ins_clust_index_entry_low(). rec_get_converted_size(), rec_get_converted_size_comp(): Allow the number of fields to be between n_core_fields and n_fields. Do not support infimum,supremum. They are never supposed to be stored in dtuple_t, because page creation nowadays uses a lower-level method for initializing them. rec_convert_dtuple_to_rec_comp(): Assign the status bits based on the number of fields. btr_cur_trim(): In an update, trim the index entry as needed. For the 'default row', handle rollback specially. For user records, omit fields that match the 'default row'. btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete(): Skip locking and adaptive hash index for the 'default row'. row_log_table_apply_convert_mrec(): Replace 'default row' values if needed. In the temporary file that is applied by row_log_table_apply(), we must identify whether the records contain the extra header for instantly added columns. For now, we will allocate an additional byte for this for ROW_T_INSERT and ROW_T_UPDATE records when the source table has been subject to instant ADD COLUMN. The ROW_T_DELETE records are fine, as they will be converted and will only contain 'core' columns (PRIMARY KEY and some system columns) that are converted from dtuple_t. rec_get_converted_size_temp(), rec_init_offsets_temp(), rec_convert_dtuple_to_temp(): Add the parameter 'status'. REC_INFO_DEFAULT_ROW = REC_INFO_MIN_REC_FLAG | REC_STATUS_COLUMNS_ADDED: An info_bits constant for distinguishing the 'default row' record. rec_comp_status_t: An enum of the status bit values. rec_leaf_format: An enum that replaces the bool parameter of rec_init_offsets_comp_ordinary().
8 years ago
MDEV-15662 Instant DROP COLUMN or changing the order of columns Allow ADD COLUMN anywhere in a table, not only adding as the last column. Allow instant DROP COLUMN and instant changing the order of columns. The added columns will always be added last in clustered index records. In new records, instantly dropped columns will be stored as NULL or empty when possible. Information about dropped and reordered columns will be written in a metadata BLOB (mblob), which is stored before the first 'user' field in the hidden metadata record at the start of the clustered index. The presence of mblob is indicated by setting the delete-mark flag in the metadata record. The metadata BLOB stores the number of clustered index fields, followed by an array of column information for each field. For dropped columns, we store the NOT NULL flag, the fixed length, and for variable-length columns, whether the maximum length exceeded 255 bytes. For non-dropped columns, we store the column position. Unlike with MDEV-11369, when a table becomes empty, it cannot be converted back to the canonical format. The reason for this is that other threads may hold cached objects such as row_prebuilt_t::ins_node that could refer to dropped or reordered index fields. For instant DROP COLUMN and ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC, we must store the n_core_null_bytes in the root page, so that the chain of node pointer records can be followed in order to reach the leftmost leaf page where the metadata record is located. If the mblob is present, we will zero-initialize the strings "infimum" and "supremum" in the root page, and use the last byte of "supremum" for storing the number of null bytes (which are allocated but useless on node pointer pages). This is necessary for btr_cur_instant_init_metadata() to be able to navigate to the mblob. If the PRIMARY KEY contains any variable-length column and some nullable columns were instantly dropped, the dict_index_t::n_nullable in the data dictionary could be smaller than it actually is in the non-leaf pages. Because of this, the non-leaf pages could use more bytes for the null flags than the data dictionary expects, and we could be reading the lengths of the variable-length columns from the wrong offset, and thus reading the child page number from wrong place. This is the result of two design mistakes that involve unnecessary storage of data: First, it is nonsense to store any data fields for the leftmost node pointer records, because the comparisons would be resolved by the MIN_REC_FLAG alone. Second, there cannot be any null fields in the clustered index node pointer fields, but we nevertheless reserve space for all the null flags. Limitations (future work): MDEV-17459 Allow instant ALTER TABLE even if FULLTEXT INDEX exists MDEV-17468 Avoid table rebuild on operations on generated columns MDEV-17494 Refuse ALGORITHM=INSTANT when the row size is too large btr_page_reorganize_low(): Preserve any metadata in the root page. Call lock_move_reorganize_page() only after restoring the "infimum" and "supremum" records, to avoid a memcmp() assertion failure. dict_col_t::DROPPED: Magic value for dict_col_t::ind. dict_col_t::clear_instant(): Renamed from dict_col_t::remove_instant(). Do not assert that the column was instantly added, because we sometimes call this unconditionally for all columns. Convert an instantly added column to a "core column". The old name remove_instant() could be mistaken to refer to "instant DROP COLUMN". dict_col_t::is_added(): Rename from dict_col_t::is_instant(). dtype_t::metadata_blob_init(): Initialize the mblob data type. dtuple_t::is_metadata(), dtuple_t::is_alter_metadata(), upd_t::is_metadata(), upd_t::is_alter_metadata(): Check if info_bits refer to a metadata record. dict_table_t::instant: Metadata about dropped or reordered columns. dict_table_t::prepare_instant(): Prepare ha_innobase_inplace_ctx::instant_table for instant ALTER TABLE. innobase_instant_try() will pass this to dict_table_t::instant_column(). On rollback, dict_table_t::rollback_instant() will be called. dict_table_t::instant_column(): Renamed from instant_add_column(). Add the parameter col_map so that columns can be reordered. Copy and adjust v_cols[] as well. dict_table_t::find(): Find an old column based on a new column number. dict_table_t::serialise_columns(), dict_table_t::deserialise_columns(): Convert the mblob. dict_index_t::instant_metadata(): Create the metadata record for instant ALTER TABLE. Invoke dict_table_t::serialise_columns(). dict_index_t::reconstruct_fields(): Invoked by dict_table_t::deserialise_columns(). dict_index_t::clear_instant_alter(): Move the fields for the dropped columns to the end, and sort the surviving index fields in ascending order of column position. ha_innobase::check_if_supported_inplace_alter(): Do not allow adding a FTS_DOC_ID column if a hidden FTS_DOC_ID column exists due to FULLTEXT INDEX. (This always required ALGORITHM=COPY.) instant_alter_column_possible(): Add a parameter for InnoDB table, to check for additional conditions, such as the maximum number of index fields. ha_innobase_inplace_ctx::first_alter_pos: The first column whose position is affected by instant ADD, DROP, or changing the order of columns. innobase_build_col_map(): Skip added virtual columns. prepare_inplace_add_virtual(): Correctly compute num_to_add_vcol. Remove some unnecessary code. Note that the call to innodb_base_col_setup() should be executed later. commit_try_norebuild(): If ctx->is_instant(), let the virtual columns be added or dropped by innobase_instant_try(). innobase_instant_try(): Fill in a zero default value for the hidden column FTS_DOC_ID (to reduce the work needed in MDEV-17459). If any columns were dropped or reordered (or added not last), delete any SYS_COLUMNS records for the following columns, and insert SYS_COLUMNS records for all subsequent stored columns as well as for all virtual columns. If any virtual column is dropped, rewrite all virtual column metadata. Use a shortcut only for adding virtual columns. This is because innobase_drop_virtual_try() assumes that the dropped virtual columns still exist in ctx->old_table. innodb_update_cols(): Renamed from innodb_update_n_cols(). innobase_add_one_virtual(), innobase_insert_sys_virtual(): Change the return type to bool, and invoke my_error() when detecting an error. innodb_insert_sys_columns(): Insert a record into SYS_COLUMNS. Refactored from innobase_add_one_virtual() and innobase_instant_add_col(). innobase_instant_add_col(): Replace the parameter dfield with type. innobase_instant_drop_cols(): Drop matching columns from SYS_COLUMNS and all columns from SYS_VIRTUAL. innobase_add_virtual_try(), innobase_drop_virtual_try(): Let the caller invoke innodb_update_cols(). innobase_rename_column_try(): Skip dropped columns. commit_cache_norebuild(): Update table->fts->doc_col. dict_mem_table_col_rename_low(): Skip dropped columns. trx_undo_rec_get_partial_row(): Skip dropped columns. trx_undo_update_rec_get_update(): Handle the metadata BLOB correctly. trx_undo_page_report_modify(): Avoid out-of-bounds access to record fields. Log metadata records consistently. Apparently, the first fields of a clustered index may be updated in an update_undo vector when the index is ID_IND of SYS_FOREIGN, as part of renaming the table during ALTER TABLE. Normally, updates of the PRIMARY KEY should be logged as delete-mark and an insert. row_undo_mod_parse_undo_rec(), row_purge_parse_undo_rec(): Use trx_undo_metadata. row_undo_mod_clust_low(): On metadata rollback, roll back the root page too. row_undo_mod_clust(): Relax an assertion. The delete-mark flag was repurposed for ALTER TABLE metadata records. row_rec_to_index_entry_impl(): Add the template parameter mblob and the optional parameter info_bits for specifying the desired new info bits. For the metadata tuple, allow conversion between the original format (ADD COLUMN only) and the generic format (with hidden BLOB). Add the optional parameter "pad" to determine whether the tuple should be padded to the index fields (on ALTER TABLE it should), or whether it should remain at its original size (on rollback). row_build_index_entry_low(): Clean up the code, removing redundant variables and conditions. For instantly dropped columns, generate a dummy value that is NULL, the empty string, or a fixed length of NUL bytes, depending on the type of the dropped column. row_upd_clust_rec_by_insert_inherit_func(): On the update of PRIMARY KEY of a record that contained a dropped column whose value was stored externally, we will be inserting a dummy NULL or empty string value to the field of the dropped column. The externally stored column would eventually be dropped when purge removes the delete-marked record for the old PRIMARY KEY value. btr_index_rec_validate(): Recognize the metadata record. btr_discard_only_page_on_level(): Preserve the generic instant ALTER TABLE metadata. btr_set_instant(): Replaces page_set_instant(). This sets a clustered index root page to the appropriate format, or upgrades from the MDEV-11369 instant ADD COLUMN to generic ALTER TABLE format. btr_cur_instant_init_low(): Read and validate the metadata BLOB page before reconstructing the dictionary information based on it. btr_cur_instant_init_metadata(): Do not read any lengths from the metadata record header before reading the BLOB. At this point, we would not actually know how many nullable fields the metadata record contains. btr_cur_instant_root_init(): Initialize n_core_null_bytes in one of two possible ways. btr_cur_trim(): Handle the mblob record. row_metadata_to_tuple(): Convert a metadata record to a data tuple, based on the new info_bits of the metadata record. btr_cur_pessimistic_update(): Invoke row_metadata_to_tuple() if needed. Invoke dtuple_convert_big_rec() for metadata records if the record is too large, or if the mblob is not yet marked as externally stored. btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete(): When the last user record is deleted, do not delete the generic instant ALTER TABLE metadata record. Only delete MDEV-11369 instant ADD COLUMN metadata records. btr_cur_optimistic_insert(): Avoid unnecessary computation of rec_size. btr_pcur_store_position(): Allow a logically empty page to contain a metadata record for generic ALTER TABLE. REC_INFO_DEFAULT_ROW_ADD: Renamed from REC_INFO_DEFAULT_ROW. This is for the old instant ADD COLUMN (MDEV-11369) only. REC_INFO_DEFAULT_ROW_ALTER: The more generic metadata record, with additional information for dropped or reordered columns. rec_info_bits_valid(): Remove. The only case when this would fail is when the record is the generic ALTER TABLE metadata record. rec_is_alter_metadata(): Check if a record is the metadata record for instant ALTER TABLE (other than ADD COLUMN). NOTE: This function must not be invoked on node pointer records, because the delete-mark flag in those records may be set (it is garbage), and then a debug assertion could fail because index->is_instant() does not necessarily hold. rec_is_add_metadata(): Check if a record is MDEV-11369 ADD COLUMN metadata record (not more generic instant ALTER TABLE). rec_get_converted_size_comp_prefix_low(): Assume that the metadata field will be stored externally. In dtuple_convert_big_rec() during the rec_get_converted_size() call, it would not be there yet. rec_get_converted_size_comp(): Replace status,fields,n_fields with tuple. rec_init_offsets_comp_ordinary(), rec_get_converted_size_comp_prefix_low(), rec_convert_dtuple_to_rec_comp(): Add template<bool mblob = false>. With mblob=true, process a record with a metadata BLOB. rec_copy_prefix_to_buf(): Assert that no fields beyond the key and system columns are being copied. Exclude the metadata BLOB field. rec_convert_dtuple_to_metadata_comp(): Convert an alter metadata tuple into a record. row_upd_index_replace_metadata(): Apply an update vector to an alter_metadata tuple. row_log_allocate(): Replace dict_index_t::is_instant() with a more appropriate condition that ignores dict_table_t::instant. Only a table on which the MDEV-11369 ADD COLUMN was performed can "lose its instantness" when it becomes empty. After instant DROP COLUMN or reordering columns, we cannot simply convert the table to the canonical format, because the data dictionary cache and all possibly existing references to it from other client connection threads would have to be adjusted. row_quiesce_write_index_fields(): Do not crash when the table contains an instantly dropped column. Thanks to Thirunarayanan Balathandayuthapani for discussing the design and implementing an initial prototype of this. Thanks to Matthias Leich for testing.
7 years ago
MDEV-15662 Instant DROP COLUMN or changing the order of columns Allow ADD COLUMN anywhere in a table, not only adding as the last column. Allow instant DROP COLUMN and instant changing the order of columns. The added columns will always be added last in clustered index records. In new records, instantly dropped columns will be stored as NULL or empty when possible. Information about dropped and reordered columns will be written in a metadata BLOB (mblob), which is stored before the first 'user' field in the hidden metadata record at the start of the clustered index. The presence of mblob is indicated by setting the delete-mark flag in the metadata record. The metadata BLOB stores the number of clustered index fields, followed by an array of column information for each field. For dropped columns, we store the NOT NULL flag, the fixed length, and for variable-length columns, whether the maximum length exceeded 255 bytes. For non-dropped columns, we store the column position. Unlike with MDEV-11369, when a table becomes empty, it cannot be converted back to the canonical format. The reason for this is that other threads may hold cached objects such as row_prebuilt_t::ins_node that could refer to dropped or reordered index fields. For instant DROP COLUMN and ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC, we must store the n_core_null_bytes in the root page, so that the chain of node pointer records can be followed in order to reach the leftmost leaf page where the metadata record is located. If the mblob is present, we will zero-initialize the strings "infimum" and "supremum" in the root page, and use the last byte of "supremum" for storing the number of null bytes (which are allocated but useless on node pointer pages). This is necessary for btr_cur_instant_init_metadata() to be able to navigate to the mblob. If the PRIMARY KEY contains any variable-length column and some nullable columns were instantly dropped, the dict_index_t::n_nullable in the data dictionary could be smaller than it actually is in the non-leaf pages. Because of this, the non-leaf pages could use more bytes for the null flags than the data dictionary expects, and we could be reading the lengths of the variable-length columns from the wrong offset, and thus reading the child page number from wrong place. This is the result of two design mistakes that involve unnecessary storage of data: First, it is nonsense to store any data fields for the leftmost node pointer records, because the comparisons would be resolved by the MIN_REC_FLAG alone. Second, there cannot be any null fields in the clustered index node pointer fields, but we nevertheless reserve space for all the null flags. Limitations (future work): MDEV-17459 Allow instant ALTER TABLE even if FULLTEXT INDEX exists MDEV-17468 Avoid table rebuild on operations on generated columns MDEV-17494 Refuse ALGORITHM=INSTANT when the row size is too large btr_page_reorganize_low(): Preserve any metadata in the root page. Call lock_move_reorganize_page() only after restoring the "infimum" and "supremum" records, to avoid a memcmp() assertion failure. dict_col_t::DROPPED: Magic value for dict_col_t::ind. dict_col_t::clear_instant(): Renamed from dict_col_t::remove_instant(). Do not assert that the column was instantly added, because we sometimes call this unconditionally for all columns. Convert an instantly added column to a "core column". The old name remove_instant() could be mistaken to refer to "instant DROP COLUMN". dict_col_t::is_added(): Rename from dict_col_t::is_instant(). dtype_t::metadata_blob_init(): Initialize the mblob data type. dtuple_t::is_metadata(), dtuple_t::is_alter_metadata(), upd_t::is_metadata(), upd_t::is_alter_metadata(): Check if info_bits refer to a metadata record. dict_table_t::instant: Metadata about dropped or reordered columns. dict_table_t::prepare_instant(): Prepare ha_innobase_inplace_ctx::instant_table for instant ALTER TABLE. innobase_instant_try() will pass this to dict_table_t::instant_column(). On rollback, dict_table_t::rollback_instant() will be called. dict_table_t::instant_column(): Renamed from instant_add_column(). Add the parameter col_map so that columns can be reordered. Copy and adjust v_cols[] as well. dict_table_t::find(): Find an old column based on a new column number. dict_table_t::serialise_columns(), dict_table_t::deserialise_columns(): Convert the mblob. dict_index_t::instant_metadata(): Create the metadata record for instant ALTER TABLE. Invoke dict_table_t::serialise_columns(). dict_index_t::reconstruct_fields(): Invoked by dict_table_t::deserialise_columns(). dict_index_t::clear_instant_alter(): Move the fields for the dropped columns to the end, and sort the surviving index fields in ascending order of column position. ha_innobase::check_if_supported_inplace_alter(): Do not allow adding a FTS_DOC_ID column if a hidden FTS_DOC_ID column exists due to FULLTEXT INDEX. (This always required ALGORITHM=COPY.) instant_alter_column_possible(): Add a parameter for InnoDB table, to check for additional conditions, such as the maximum number of index fields. ha_innobase_inplace_ctx::first_alter_pos: The first column whose position is affected by instant ADD, DROP, or changing the order of columns. innobase_build_col_map(): Skip added virtual columns. prepare_inplace_add_virtual(): Correctly compute num_to_add_vcol. Remove some unnecessary code. Note that the call to innodb_base_col_setup() should be executed later. commit_try_norebuild(): If ctx->is_instant(), let the virtual columns be added or dropped by innobase_instant_try(). innobase_instant_try(): Fill in a zero default value for the hidden column FTS_DOC_ID (to reduce the work needed in MDEV-17459). If any columns were dropped or reordered (or added not last), delete any SYS_COLUMNS records for the following columns, and insert SYS_COLUMNS records for all subsequent stored columns as well as for all virtual columns. If any virtual column is dropped, rewrite all virtual column metadata. Use a shortcut only for adding virtual columns. This is because innobase_drop_virtual_try() assumes that the dropped virtual columns still exist in ctx->old_table. innodb_update_cols(): Renamed from innodb_update_n_cols(). innobase_add_one_virtual(), innobase_insert_sys_virtual(): Change the return type to bool, and invoke my_error() when detecting an error. innodb_insert_sys_columns(): Insert a record into SYS_COLUMNS. Refactored from innobase_add_one_virtual() and innobase_instant_add_col(). innobase_instant_add_col(): Replace the parameter dfield with type. innobase_instant_drop_cols(): Drop matching columns from SYS_COLUMNS and all columns from SYS_VIRTUAL. innobase_add_virtual_try(), innobase_drop_virtual_try(): Let the caller invoke innodb_update_cols(). innobase_rename_column_try(): Skip dropped columns. commit_cache_norebuild(): Update table->fts->doc_col. dict_mem_table_col_rename_low(): Skip dropped columns. trx_undo_rec_get_partial_row(): Skip dropped columns. trx_undo_update_rec_get_update(): Handle the metadata BLOB correctly. trx_undo_page_report_modify(): Avoid out-of-bounds access to record fields. Log metadata records consistently. Apparently, the first fields of a clustered index may be updated in an update_undo vector when the index is ID_IND of SYS_FOREIGN, as part of renaming the table during ALTER TABLE. Normally, updates of the PRIMARY KEY should be logged as delete-mark and an insert. row_undo_mod_parse_undo_rec(), row_purge_parse_undo_rec(): Use trx_undo_metadata. row_undo_mod_clust_low(): On metadata rollback, roll back the root page too. row_undo_mod_clust(): Relax an assertion. The delete-mark flag was repurposed for ALTER TABLE metadata records. row_rec_to_index_entry_impl(): Add the template parameter mblob and the optional parameter info_bits for specifying the desired new info bits. For the metadata tuple, allow conversion between the original format (ADD COLUMN only) and the generic format (with hidden BLOB). Add the optional parameter "pad" to determine whether the tuple should be padded to the index fields (on ALTER TABLE it should), or whether it should remain at its original size (on rollback). row_build_index_entry_low(): Clean up the code, removing redundant variables and conditions. For instantly dropped columns, generate a dummy value that is NULL, the empty string, or a fixed length of NUL bytes, depending on the type of the dropped column. row_upd_clust_rec_by_insert_inherit_func(): On the update of PRIMARY KEY of a record that contained a dropped column whose value was stored externally, we will be inserting a dummy NULL or empty string value to the field of the dropped column. The externally stored column would eventually be dropped when purge removes the delete-marked record for the old PRIMARY KEY value. btr_index_rec_validate(): Recognize the metadata record. btr_discard_only_page_on_level(): Preserve the generic instant ALTER TABLE metadata. btr_set_instant(): Replaces page_set_instant(). This sets a clustered index root page to the appropriate format, or upgrades from the MDEV-11369 instant ADD COLUMN to generic ALTER TABLE format. btr_cur_instant_init_low(): Read and validate the metadata BLOB page before reconstructing the dictionary information based on it. btr_cur_instant_init_metadata(): Do not read any lengths from the metadata record header before reading the BLOB. At this point, we would not actually know how many nullable fields the metadata record contains. btr_cur_instant_root_init(): Initialize n_core_null_bytes in one of two possible ways. btr_cur_trim(): Handle the mblob record. row_metadata_to_tuple(): Convert a metadata record to a data tuple, based on the new info_bits of the metadata record. btr_cur_pessimistic_update(): Invoke row_metadata_to_tuple() if needed. Invoke dtuple_convert_big_rec() for metadata records if the record is too large, or if the mblob is not yet marked as externally stored. btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete(): When the last user record is deleted, do not delete the generic instant ALTER TABLE metadata record. Only delete MDEV-11369 instant ADD COLUMN metadata records. btr_cur_optimistic_insert(): Avoid unnecessary computation of rec_size. btr_pcur_store_position(): Allow a logically empty page to contain a metadata record for generic ALTER TABLE. REC_INFO_DEFAULT_ROW_ADD: Renamed from REC_INFO_DEFAULT_ROW. This is for the old instant ADD COLUMN (MDEV-11369) only. REC_INFO_DEFAULT_ROW_ALTER: The more generic metadata record, with additional information for dropped or reordered columns. rec_info_bits_valid(): Remove. The only case when this would fail is when the record is the generic ALTER TABLE metadata record. rec_is_alter_metadata(): Check if a record is the metadata record for instant ALTER TABLE (other than ADD COLUMN). NOTE: This function must not be invoked on node pointer records, because the delete-mark flag in those records may be set (it is garbage), and then a debug assertion could fail because index->is_instant() does not necessarily hold. rec_is_add_metadata(): Check if a record is MDEV-11369 ADD COLUMN metadata record (not more generic instant ALTER TABLE). rec_get_converted_size_comp_prefix_low(): Assume that the metadata field will be stored externally. In dtuple_convert_big_rec() during the rec_get_converted_size() call, it would not be there yet. rec_get_converted_size_comp(): Replace status,fields,n_fields with tuple. rec_init_offsets_comp_ordinary(), rec_get_converted_size_comp_prefix_low(), rec_convert_dtuple_to_rec_comp(): Add template<bool mblob = false>. With mblob=true, process a record with a metadata BLOB. rec_copy_prefix_to_buf(): Assert that no fields beyond the key and system columns are being copied. Exclude the metadata BLOB field. rec_convert_dtuple_to_metadata_comp(): Convert an alter metadata tuple into a record. row_upd_index_replace_metadata(): Apply an update vector to an alter_metadata tuple. row_log_allocate(): Replace dict_index_t::is_instant() with a more appropriate condition that ignores dict_table_t::instant. Only a table on which the MDEV-11369 ADD COLUMN was performed can "lose its instantness" when it becomes empty. After instant DROP COLUMN or reordering columns, we cannot simply convert the table to the canonical format, because the data dictionary cache and all possibly existing references to it from other client connection threads would have to be adjusted. row_quiesce_write_index_fields(): Do not crash when the table contains an instantly dropped column. Thanks to Thirunarayanan Balathandayuthapani for discussing the design and implementing an initial prototype of this. Thanks to Matthias Leich for testing.
7 years ago
MDEV-24971 InnoDB access freed virtual column after rollback of secondary index Problem: ======== InnoDB fails to clean the index stub if it fails to add the virtual index which contains new virtual column. But it clears the newly virtual column from index in clear_added_indexes() during inplace_alter_table. On commit, InnoDB evicts and reload the table. In case of rollback, it doesn't happen. InnoDB clears the ABORTED index while opening the table or doing the DDL. In the mean time, InnoDB can access the dropped virtual index columns while creating prebuilt or rollback of concurrent DML. Solution: ========== (1) InnoDB should maintain newly added virtual column while rollbacking the newly added virtual index. (2) InnoDB must not defer the index removal if the alter table is executed with LOCK=EXCLUSIVE. (3) For LOCK=SHARED, InnoDB should check whether the table has any other transaction lock other than alter transaction before deferring the index stub. Replaced has_new_v_col with dict_add_vcol_info in dict_index_t to indicate whether the index has any new virtual column. dict_index_t::has_new_v_col(): Returns whether the index has newly added virtual column, it doesn't say which columns are newly added virtual column ha_innobase_inplace_ctx::is_new_vcol(): Return whether the given column is added as a part of the current alter. ha_innobase_inplace_ctx::clean_new_vcol_index(): Copy the newly added virtual column to new_vcol_info in dict_index_t. Replace the column in the index fields with virtual column stored in new_vcol_info. dict_index_t::assign_new_v_col(): Store the number of virtual column added in index as a part of alter table. dict_index_t::get_n_new_vcol(): Get the number of newly added virtual column dict_index_t::assign_drop_v_col(): Allocate the memory for adding new virtual column in new_vcol_info. dict_index_t::add_drop_v_col(): Add the newly added virtual column in new_vcol_info. dict_table_t::has_lock_for_other_trx(): Whether the table has any other transaction lock than given transaction. row_merge_drop_indexes(): Add parameter alter_trx and check whether the table has any other lock than alter transaction.
5 years ago
MDEV-24971 InnoDB access freed virtual column after rollback of secondary index Problem: ======== InnoDB fails to clean the index stub if it fails to add the virtual index which contains new virtual column. But it clears the newly virtual column from index in clear_added_indexes() during inplace_alter_table. On commit, InnoDB evicts and reload the table. In case of rollback, it doesn't happen. InnoDB clears the ABORTED index while opening the table or doing the DDL. In the mean time, InnoDB can access the dropped virtual index columns while creating prebuilt or rollback of concurrent DML. Solution: ========== (1) InnoDB should maintain newly added virtual column while rollbacking the newly added virtual index. (2) InnoDB must not defer the index removal if the alter table is executed with LOCK=EXCLUSIVE. (3) For LOCK=SHARED, InnoDB should check whether the table has any other transaction lock other than alter transaction before deferring the index stub. Replaced has_new_v_col with dict_add_vcol_info in dict_index_t to indicate whether the index has any new virtual column. dict_index_t::has_new_v_col(): Returns whether the index has newly added virtual column, it doesn't say which columns are newly added virtual column ha_innobase_inplace_ctx::is_new_vcol(): Return whether the given column is added as a part of the current alter. ha_innobase_inplace_ctx::clean_new_vcol_index(): Copy the newly added virtual column to new_vcol_info in dict_index_t. Replace the column in the index fields with virtual column stored in new_vcol_info. dict_index_t::assign_new_v_col(): Store the number of virtual column added in index as a part of alter table. dict_index_t::get_n_new_vcol(): Get the number of newly added virtual column dict_index_t::assign_drop_v_col(): Allocate the memory for adding new virtual column in new_vcol_info. dict_index_t::add_drop_v_col(): Add the newly added virtual column in new_vcol_info. dict_table_t::has_lock_for_other_trx(): Whether the table has any other transaction lock than given transaction. row_merge_drop_indexes(): Add parameter alter_trx and check whether the table has any other lock than alter transaction.
5 years ago
MDEV-24971 InnoDB access freed virtual column after rollback of secondary index Problem: ======== InnoDB fails to clean the index stub if it fails to add the virtual index which contains new virtual column. But it clears the newly virtual column from index in clear_added_indexes() during inplace_alter_table. On commit, InnoDB evicts and reload the table. In case of rollback, it doesn't happen. InnoDB clears the ABORTED index while opening the table or doing the DDL. In the mean time, InnoDB can access the dropped virtual index columns while creating prebuilt or rollback of concurrent DML. Solution: ========== (1) InnoDB should maintain newly added virtual column while rollbacking the newly added virtual index. (2) InnoDB must not defer the index removal if the alter table is executed with LOCK=EXCLUSIVE. (3) For LOCK=SHARED, InnoDB should check whether the table has any other transaction lock other than alter transaction before deferring the index stub. Replaced has_new_v_col with dict_add_vcol_info in dict_index_t to indicate whether the index has any new virtual column. dict_index_t::has_new_v_col(): Returns whether the index has newly added virtual column, it doesn't say which columns are newly added virtual column ha_innobase_inplace_ctx::is_new_vcol(): Return whether the given column is added as a part of the current alter. ha_innobase_inplace_ctx::clean_new_vcol_index(): Copy the newly added virtual column to new_vcol_info in dict_index_t. Replace the column in the index fields with virtual column stored in new_vcol_info. dict_index_t::assign_new_v_col(): Store the number of virtual column added in index as a part of alter table. dict_index_t::get_n_new_vcol(): Get the number of newly added virtual column dict_index_t::assign_drop_v_col(): Allocate the memory for adding new virtual column in new_vcol_info. dict_index_t::add_drop_v_col(): Add the newly added virtual column in new_vcol_info. dict_table_t::has_lock_for_other_trx(): Whether the table has any other transaction lock than given transaction. row_merge_drop_indexes(): Add parameter alter_trx and check whether the table has any other lock than alter transaction.
5 years ago
MDEV-22456 Dropping the adaptive hash index may cause DDL to lock up InnoDB If the InnoDB buffer pool contains many pages for a table or index that is being dropped or rebuilt, and if many of such pages are pointed to by the adaptive hash index, dropping the adaptive hash index may consume a lot of time. The time-consuming operation of dropping the adaptive hash index entries is being executed while the InnoDB data dictionary cache dict_sys is exclusively locked. It is not actually necessary to drop all adaptive hash index entries at the time a table or index is being dropped or rebuilt. We can let the LRU replacement policy of the buffer pool take care of this gradually. For this to work, we must detach the dict_table_t and dict_index_t objects from the main dict_sys cache, and once the last adaptive hash index entry for the detached table is removed (when the garbage page is evicted from the buffer pool) we can free the dict_table_t and dict_index_t object. Related to this, in MDEV-16283, we made ALTER TABLE...DISCARD TABLESPACE skip both the buffer pool eviction and the drop of the adaptive hash index. We shifted the burden to ALTER TABLE...IMPORT TABLESPACE or DROP TABLE. We can remove the eviction from DROP TABLE. We must retain the eviction in the ALTER TABLE...IMPORT TABLESPACE code path, so that in case the discarded table is being re-imported with the same tablespace identifier, the fresh data from the imported tablespace will replace any stale pages in the buffer pool. rpl.rpl_failed_drop_tbl_binlog: Remove the test. DROP TABLE can no longer be interrupted inside InnoDB. fseg_free_page(), fseg_free_step(), fseg_free_step_not_header(), fseg_free_page_low(), fseg_free_extent(): Remove the parameter that specifies whether the adaptive hash index should be dropped. btr_search_lazy_free(): Lazily free an index when the last reference to it is dropped from the adaptive hash index. buf_pool_clear_hash_index(): Declare static, and move to the same compilation unit with the bulk of the adaptive hash index code. dict_index_t::clone(), dict_index_t::clone_if_needed(): Clone an index that is being rebuilt while adaptive hash index entries exist. The original index will be inserted into dict_table_t::freed_indexes and dict_index_t::set_freed() will be called. dict_index_t::set_freed(), dict_index_t::freed(): Note that or check whether the index has been freed. We will use the impossible page number 1 to denote this condition. dict_index_t::n_ahi_pages(): Replaces btr_search_info_get_ref_count(). dict_index_t::detach_columns(): Move the assignment n_fields=0 to ha_innobase_inplace_ctx::clear_added_indexes(). We must have access to the columns when freeing the adaptive hash index. Note: dict_table_t::v_cols[] will remain valid. If virtual columns are dropped or added, the table definition will be reloaded in ha_innobase::commit_inplace_alter_table(). buf_page_mtr_lock(): Drop a stale adaptive hash index if needed. We will also reduce the number of btr_get_search_latch() calls and enclose some more code inside #ifdef BTR_CUR_HASH_ADAPT in order to benefit cmake -DWITH_INNODB_AHI=OFF.
5 years ago
MDEV-22867 Assertion instant.n_core_fields == n_core_fields failed This is a race condition where a table on which a 10.3-style instant ADD COLUMN is emptied during the execution of ALTER TABLE ... DROP COLUMN ..., DROP INDEX ..., ALGORITHM=NOCOPY. In commit 2c4844c9e76427525e8c39a2d72686085efe89c3 the function instant_metadata_lock() would prevent this race condition. But, it would also hold a page latch on the leftmost leaf page of clustered index for the duration of a possible DROP INDEX operation. The race could be fixed by restoring the function instant_metadata_lock() that was removed in commit ea37b144094a0c2ebfc6774047fd473c1b2a8658 but it would be more future-proof to prevent the dict_index_t::clear_instant_add() call from being issued at all. We at some point support DROP COLUMN ..., ADD INDEX ..., ALGORITHM=NOCOPY and that would spend a non-trivial amount of execution time in ha_innobase::inplace_alter(), making a server hang possible. Currently this is not supported and our added test case will notice when the support is introduced. dict_index_t::must_avoid_clear_instant_add(): Determine if a call to clear_instant_add() must be avoided. btr_discard_only_page_on_level(): Preserve the metadata record if must_avoid_clear_instant_add() holds. btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete(): Do not remove the metadata record even if the table becomes empty but must_avoid_clear_instant_add() holds. btr_pcur_store_position(): Relax a debug assertion. This is joint work with Thirunarayanan Balathandayuthapani.
5 years ago
11 years ago
11 years ago
11 years ago
11 years ago
11 years ago
11 years ago
11 years ago
11 years ago
11 years ago
MDEV-21907: InnoDB: Enable -Wconversion on clang and GCC The -Wconversion in GCC seems to be stricter than in clang. GCC at least since version 4.4.7 issues truncation warnings for assignments to bitfields, while clang 10 appears to only issue warnings when the sizes in bytes rounded to the nearest integer powers of 2 are different. Before GCC 10.0.0, -Wconversion required more casts and would not allow some operations, such as x<<=1 or x+=1 on a data type that is narrower than int. GCC 5 (but not GCC 4, GCC 6, or any later version) is complaining about x|=y even when x and y are compatible types that are narrower than int. Hence, we must rewrite some x|=y as x=static_cast<byte>(x|y) or similar, or we must disable -Wconversion. In GCC 6 and later, the warning for assigning wider to bitfields that are narrower than 8, 16, or 32 bits can be suppressed by applying a bitwise & with the exact bitmask of the bitfield. For older GCC, we must disable -Wconversion for GCC 4 or 5 in such cases. The bitwise negation operator appears to promote short integers to a wider type, and hence we must add explicit truncation casts around them. Microsoft Visual C does not allow a static_cast to truncate a constant, such as static_cast<byte>(1) truncating int. Hence, we will use the constructor-style cast byte(~1) for such cases. This has been tested at least with GCC 4.8.5, 5.4.0, 7.4.0, 9.2.1, 10.0.0, clang 9.0.1, 10.0.0, and MSVC 14.22.27905 (Microsoft Visual Studio 2019) on 64-bit and 32-bit targets (IA-32, AMD64, POWER 8, POWER 9, ARMv8).
6 years ago
MDEV-15662 Instant DROP COLUMN or changing the order of columns Allow ADD COLUMN anywhere in a table, not only adding as the last column. Allow instant DROP COLUMN and instant changing the order of columns. The added columns will always be added last in clustered index records. In new records, instantly dropped columns will be stored as NULL or empty when possible. Information about dropped and reordered columns will be written in a metadata BLOB (mblob), which is stored before the first 'user' field in the hidden metadata record at the start of the clustered index. The presence of mblob is indicated by setting the delete-mark flag in the metadata record. The metadata BLOB stores the number of clustered index fields, followed by an array of column information for each field. For dropped columns, we store the NOT NULL flag, the fixed length, and for variable-length columns, whether the maximum length exceeded 255 bytes. For non-dropped columns, we store the column position. Unlike with MDEV-11369, when a table becomes empty, it cannot be converted back to the canonical format. The reason for this is that other threads may hold cached objects such as row_prebuilt_t::ins_node that could refer to dropped or reordered index fields. For instant DROP COLUMN and ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC, we must store the n_core_null_bytes in the root page, so that the chain of node pointer records can be followed in order to reach the leftmost leaf page where the metadata record is located. If the mblob is present, we will zero-initialize the strings "infimum" and "supremum" in the root page, and use the last byte of "supremum" for storing the number of null bytes (which are allocated but useless on node pointer pages). This is necessary for btr_cur_instant_init_metadata() to be able to navigate to the mblob. If the PRIMARY KEY contains any variable-length column and some nullable columns were instantly dropped, the dict_index_t::n_nullable in the data dictionary could be smaller than it actually is in the non-leaf pages. Because of this, the non-leaf pages could use more bytes for the null flags than the data dictionary expects, and we could be reading the lengths of the variable-length columns from the wrong offset, and thus reading the child page number from wrong place. This is the result of two design mistakes that involve unnecessary storage of data: First, it is nonsense to store any data fields for the leftmost node pointer records, because the comparisons would be resolved by the MIN_REC_FLAG alone. Second, there cannot be any null fields in the clustered index node pointer fields, but we nevertheless reserve space for all the null flags. Limitations (future work): MDEV-17459 Allow instant ALTER TABLE even if FULLTEXT INDEX exists MDEV-17468 Avoid table rebuild on operations on generated columns MDEV-17494 Refuse ALGORITHM=INSTANT when the row size is too large btr_page_reorganize_low(): Preserve any metadata in the root page. Call lock_move_reorganize_page() only after restoring the "infimum" and "supremum" records, to avoid a memcmp() assertion failure. dict_col_t::DROPPED: Magic value for dict_col_t::ind. dict_col_t::clear_instant(): Renamed from dict_col_t::remove_instant(). Do not assert that the column was instantly added, because we sometimes call this unconditionally for all columns. Convert an instantly added column to a "core column". The old name remove_instant() could be mistaken to refer to "instant DROP COLUMN". dict_col_t::is_added(): Rename from dict_col_t::is_instant(). dtype_t::metadata_blob_init(): Initialize the mblob data type. dtuple_t::is_metadata(), dtuple_t::is_alter_metadata(), upd_t::is_metadata(), upd_t::is_alter_metadata(): Check if info_bits refer to a metadata record. dict_table_t::instant: Metadata about dropped or reordered columns. dict_table_t::prepare_instant(): Prepare ha_innobase_inplace_ctx::instant_table for instant ALTER TABLE. innobase_instant_try() will pass this to dict_table_t::instant_column(). On rollback, dict_table_t::rollback_instant() will be called. dict_table_t::instant_column(): Renamed from instant_add_column(). Add the parameter col_map so that columns can be reordered. Copy and adjust v_cols[] as well. dict_table_t::find(): Find an old column based on a new column number. dict_table_t::serialise_columns(), dict_table_t::deserialise_columns(): Convert the mblob. dict_index_t::instant_metadata(): Create the metadata record for instant ALTER TABLE. Invoke dict_table_t::serialise_columns(). dict_index_t::reconstruct_fields(): Invoked by dict_table_t::deserialise_columns(). dict_index_t::clear_instant_alter(): Move the fields for the dropped columns to the end, and sort the surviving index fields in ascending order of column position. ha_innobase::check_if_supported_inplace_alter(): Do not allow adding a FTS_DOC_ID column if a hidden FTS_DOC_ID column exists due to FULLTEXT INDEX. (This always required ALGORITHM=COPY.) instant_alter_column_possible(): Add a parameter for InnoDB table, to check for additional conditions, such as the maximum number of index fields. ha_innobase_inplace_ctx::first_alter_pos: The first column whose position is affected by instant ADD, DROP, or changing the order of columns. innobase_build_col_map(): Skip added virtual columns. prepare_inplace_add_virtual(): Correctly compute num_to_add_vcol. Remove some unnecessary code. Note that the call to innodb_base_col_setup() should be executed later. commit_try_norebuild(): If ctx->is_instant(), let the virtual columns be added or dropped by innobase_instant_try(). innobase_instant_try(): Fill in a zero default value for the hidden column FTS_DOC_ID (to reduce the work needed in MDEV-17459). If any columns were dropped or reordered (or added not last), delete any SYS_COLUMNS records for the following columns, and insert SYS_COLUMNS records for all subsequent stored columns as well as for all virtual columns. If any virtual column is dropped, rewrite all virtual column metadata. Use a shortcut only for adding virtual columns. This is because innobase_drop_virtual_try() assumes that the dropped virtual columns still exist in ctx->old_table. innodb_update_cols(): Renamed from innodb_update_n_cols(). innobase_add_one_virtual(), innobase_insert_sys_virtual(): Change the return type to bool, and invoke my_error() when detecting an error. innodb_insert_sys_columns(): Insert a record into SYS_COLUMNS. Refactored from innobase_add_one_virtual() and innobase_instant_add_col(). innobase_instant_add_col(): Replace the parameter dfield with type. innobase_instant_drop_cols(): Drop matching columns from SYS_COLUMNS and all columns from SYS_VIRTUAL. innobase_add_virtual_try(), innobase_drop_virtual_try(): Let the caller invoke innodb_update_cols(). innobase_rename_column_try(): Skip dropped columns. commit_cache_norebuild(): Update table->fts->doc_col. dict_mem_table_col_rename_low(): Skip dropped columns. trx_undo_rec_get_partial_row(): Skip dropped columns. trx_undo_update_rec_get_update(): Handle the metadata BLOB correctly. trx_undo_page_report_modify(): Avoid out-of-bounds access to record fields. Log metadata records consistently. Apparently, the first fields of a clustered index may be updated in an update_undo vector when the index is ID_IND of SYS_FOREIGN, as part of renaming the table during ALTER TABLE. Normally, updates of the PRIMARY KEY should be logged as delete-mark and an insert. row_undo_mod_parse_undo_rec(), row_purge_parse_undo_rec(): Use trx_undo_metadata. row_undo_mod_clust_low(): On metadata rollback, roll back the root page too. row_undo_mod_clust(): Relax an assertion. The delete-mark flag was repurposed for ALTER TABLE metadata records. row_rec_to_index_entry_impl(): Add the template parameter mblob and the optional parameter info_bits for specifying the desired new info bits. For the metadata tuple, allow conversion between the original format (ADD COLUMN only) and the generic format (with hidden BLOB). Add the optional parameter "pad" to determine whether the tuple should be padded to the index fields (on ALTER TABLE it should), or whether it should remain at its original size (on rollback). row_build_index_entry_low(): Clean up the code, removing redundant variables and conditions. For instantly dropped columns, generate a dummy value that is NULL, the empty string, or a fixed length of NUL bytes, depending on the type of the dropped column. row_upd_clust_rec_by_insert_inherit_func(): On the update of PRIMARY KEY of a record that contained a dropped column whose value was stored externally, we will be inserting a dummy NULL or empty string value to the field of the dropped column. The externally stored column would eventually be dropped when purge removes the delete-marked record for the old PRIMARY KEY value. btr_index_rec_validate(): Recognize the metadata record. btr_discard_only_page_on_level(): Preserve the generic instant ALTER TABLE metadata. btr_set_instant(): Replaces page_set_instant(). This sets a clustered index root page to the appropriate format, or upgrades from the MDEV-11369 instant ADD COLUMN to generic ALTER TABLE format. btr_cur_instant_init_low(): Read and validate the metadata BLOB page before reconstructing the dictionary information based on it. btr_cur_instant_init_metadata(): Do not read any lengths from the metadata record header before reading the BLOB. At this point, we would not actually know how many nullable fields the metadata record contains. btr_cur_instant_root_init(): Initialize n_core_null_bytes in one of two possible ways. btr_cur_trim(): Handle the mblob record. row_metadata_to_tuple(): Convert a metadata record to a data tuple, based on the new info_bits of the metadata record. btr_cur_pessimistic_update(): Invoke row_metadata_to_tuple() if needed. Invoke dtuple_convert_big_rec() for metadata records if the record is too large, or if the mblob is not yet marked as externally stored. btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete(): When the last user record is deleted, do not delete the generic instant ALTER TABLE metadata record. Only delete MDEV-11369 instant ADD COLUMN metadata records. btr_cur_optimistic_insert(): Avoid unnecessary computation of rec_size. btr_pcur_store_position(): Allow a logically empty page to contain a metadata record for generic ALTER TABLE. REC_INFO_DEFAULT_ROW_ADD: Renamed from REC_INFO_DEFAULT_ROW. This is for the old instant ADD COLUMN (MDEV-11369) only. REC_INFO_DEFAULT_ROW_ALTER: The more generic metadata record, with additional information for dropped or reordered columns. rec_info_bits_valid(): Remove. The only case when this would fail is when the record is the generic ALTER TABLE metadata record. rec_is_alter_metadata(): Check if a record is the metadata record for instant ALTER TABLE (other than ADD COLUMN). NOTE: This function must not be invoked on node pointer records, because the delete-mark flag in those records may be set (it is garbage), and then a debug assertion could fail because index->is_instant() does not necessarily hold. rec_is_add_metadata(): Check if a record is MDEV-11369 ADD COLUMN metadata record (not more generic instant ALTER TABLE). rec_get_converted_size_comp_prefix_low(): Assume that the metadata field will be stored externally. In dtuple_convert_big_rec() during the rec_get_converted_size() call, it would not be there yet. rec_get_converted_size_comp(): Replace status,fields,n_fields with tuple. rec_init_offsets_comp_ordinary(), rec_get_converted_size_comp_prefix_low(), rec_convert_dtuple_to_rec_comp(): Add template<bool mblob = false>. With mblob=true, process a record with a metadata BLOB. rec_copy_prefix_to_buf(): Assert that no fields beyond the key and system columns are being copied. Exclude the metadata BLOB field. rec_convert_dtuple_to_metadata_comp(): Convert an alter metadata tuple into a record. row_upd_index_replace_metadata(): Apply an update vector to an alter_metadata tuple. row_log_allocate(): Replace dict_index_t::is_instant() with a more appropriate condition that ignores dict_table_t::instant. Only a table on which the MDEV-11369 ADD COLUMN was performed can "lose its instantness" when it becomes empty. After instant DROP COLUMN or reordering columns, we cannot simply convert the table to the canonical format, because the data dictionary cache and all possibly existing references to it from other client connection threads would have to be adjusted. row_quiesce_write_index_fields(): Do not crash when the table contains an instantly dropped column. Thanks to Thirunarayanan Balathandayuthapani for discussing the design and implementing an initial prototype of this. Thanks to Matthias Leich for testing.
7 years ago
MDEV-15662 Instant DROP COLUMN or changing the order of columns Allow ADD COLUMN anywhere in a table, not only adding as the last column. Allow instant DROP COLUMN and instant changing the order of columns. The added columns will always be added last in clustered index records. In new records, instantly dropped columns will be stored as NULL or empty when possible. Information about dropped and reordered columns will be written in a metadata BLOB (mblob), which is stored before the first 'user' field in the hidden metadata record at the start of the clustered index. The presence of mblob is indicated by setting the delete-mark flag in the metadata record. The metadata BLOB stores the number of clustered index fields, followed by an array of column information for each field. For dropped columns, we store the NOT NULL flag, the fixed length, and for variable-length columns, whether the maximum length exceeded 255 bytes. For non-dropped columns, we store the column position. Unlike with MDEV-11369, when a table becomes empty, it cannot be converted back to the canonical format. The reason for this is that other threads may hold cached objects such as row_prebuilt_t::ins_node that could refer to dropped or reordered index fields. For instant DROP COLUMN and ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC, we must store the n_core_null_bytes in the root page, so that the chain of node pointer records can be followed in order to reach the leftmost leaf page where the metadata record is located. If the mblob is present, we will zero-initialize the strings "infimum" and "supremum" in the root page, and use the last byte of "supremum" for storing the number of null bytes (which are allocated but useless on node pointer pages). This is necessary for btr_cur_instant_init_metadata() to be able to navigate to the mblob. If the PRIMARY KEY contains any variable-length column and some nullable columns were instantly dropped, the dict_index_t::n_nullable in the data dictionary could be smaller than it actually is in the non-leaf pages. Because of this, the non-leaf pages could use more bytes for the null flags than the data dictionary expects, and we could be reading the lengths of the variable-length columns from the wrong offset, and thus reading the child page number from wrong place. This is the result of two design mistakes that involve unnecessary storage of data: First, it is nonsense to store any data fields for the leftmost node pointer records, because the comparisons would be resolved by the MIN_REC_FLAG alone. Second, there cannot be any null fields in the clustered index node pointer fields, but we nevertheless reserve space for all the null flags. Limitations (future work): MDEV-17459 Allow instant ALTER TABLE even if FULLTEXT INDEX exists MDEV-17468 Avoid table rebuild on operations on generated columns MDEV-17494 Refuse ALGORITHM=INSTANT when the row size is too large btr_page_reorganize_low(): Preserve any metadata in the root page. Call lock_move_reorganize_page() only after restoring the "infimum" and "supremum" records, to avoid a memcmp() assertion failure. dict_col_t::DROPPED: Magic value for dict_col_t::ind. dict_col_t::clear_instant(): Renamed from dict_col_t::remove_instant(). Do not assert that the column was instantly added, because we sometimes call this unconditionally for all columns. Convert an instantly added column to a "core column". The old name remove_instant() could be mistaken to refer to "instant DROP COLUMN". dict_col_t::is_added(): Rename from dict_col_t::is_instant(). dtype_t::metadata_blob_init(): Initialize the mblob data type. dtuple_t::is_metadata(), dtuple_t::is_alter_metadata(), upd_t::is_metadata(), upd_t::is_alter_metadata(): Check if info_bits refer to a metadata record. dict_table_t::instant: Metadata about dropped or reordered columns. dict_table_t::prepare_instant(): Prepare ha_innobase_inplace_ctx::instant_table for instant ALTER TABLE. innobase_instant_try() will pass this to dict_table_t::instant_column(). On rollback, dict_table_t::rollback_instant() will be called. dict_table_t::instant_column(): Renamed from instant_add_column(). Add the parameter col_map so that columns can be reordered. Copy and adjust v_cols[] as well. dict_table_t::find(): Find an old column based on a new column number. dict_table_t::serialise_columns(), dict_table_t::deserialise_columns(): Convert the mblob. dict_index_t::instant_metadata(): Create the metadata record for instant ALTER TABLE. Invoke dict_table_t::serialise_columns(). dict_index_t::reconstruct_fields(): Invoked by dict_table_t::deserialise_columns(). dict_index_t::clear_instant_alter(): Move the fields for the dropped columns to the end, and sort the surviving index fields in ascending order of column position. ha_innobase::check_if_supported_inplace_alter(): Do not allow adding a FTS_DOC_ID column if a hidden FTS_DOC_ID column exists due to FULLTEXT INDEX. (This always required ALGORITHM=COPY.) instant_alter_column_possible(): Add a parameter for InnoDB table, to check for additional conditions, such as the maximum number of index fields. ha_innobase_inplace_ctx::first_alter_pos: The first column whose position is affected by instant ADD, DROP, or changing the order of columns. innobase_build_col_map(): Skip added virtual columns. prepare_inplace_add_virtual(): Correctly compute num_to_add_vcol. Remove some unnecessary code. Note that the call to innodb_base_col_setup() should be executed later. commit_try_norebuild(): If ctx->is_instant(), let the virtual columns be added or dropped by innobase_instant_try(). innobase_instant_try(): Fill in a zero default value for the hidden column FTS_DOC_ID (to reduce the work needed in MDEV-17459). If any columns were dropped or reordered (or added not last), delete any SYS_COLUMNS records for the following columns, and insert SYS_COLUMNS records for all subsequent stored columns as well as for all virtual columns. If any virtual column is dropped, rewrite all virtual column metadata. Use a shortcut only for adding virtual columns. This is because innobase_drop_virtual_try() assumes that the dropped virtual columns still exist in ctx->old_table. innodb_update_cols(): Renamed from innodb_update_n_cols(). innobase_add_one_virtual(), innobase_insert_sys_virtual(): Change the return type to bool, and invoke my_error() when detecting an error. innodb_insert_sys_columns(): Insert a record into SYS_COLUMNS. Refactored from innobase_add_one_virtual() and innobase_instant_add_col(). innobase_instant_add_col(): Replace the parameter dfield with type. innobase_instant_drop_cols(): Drop matching columns from SYS_COLUMNS and all columns from SYS_VIRTUAL. innobase_add_virtual_try(), innobase_drop_virtual_try(): Let the caller invoke innodb_update_cols(). innobase_rename_column_try(): Skip dropped columns. commit_cache_norebuild(): Update table->fts->doc_col. dict_mem_table_col_rename_low(): Skip dropped columns. trx_undo_rec_get_partial_row(): Skip dropped columns. trx_undo_update_rec_get_update(): Handle the metadata BLOB correctly. trx_undo_page_report_modify(): Avoid out-of-bounds access to record fields. Log metadata records consistently. Apparently, the first fields of a clustered index may be updated in an update_undo vector when the index is ID_IND of SYS_FOREIGN, as part of renaming the table during ALTER TABLE. Normally, updates of the PRIMARY KEY should be logged as delete-mark and an insert. row_undo_mod_parse_undo_rec(), row_purge_parse_undo_rec(): Use trx_undo_metadata. row_undo_mod_clust_low(): On metadata rollback, roll back the root page too. row_undo_mod_clust(): Relax an assertion. The delete-mark flag was repurposed for ALTER TABLE metadata records. row_rec_to_index_entry_impl(): Add the template parameter mblob and the optional parameter info_bits for specifying the desired new info bits. For the metadata tuple, allow conversion between the original format (ADD COLUMN only) and the generic format (with hidden BLOB). Add the optional parameter "pad" to determine whether the tuple should be padded to the index fields (on ALTER TABLE it should), or whether it should remain at its original size (on rollback). row_build_index_entry_low(): Clean up the code, removing redundant variables and conditions. For instantly dropped columns, generate a dummy value that is NULL, the empty string, or a fixed length of NUL bytes, depending on the type of the dropped column. row_upd_clust_rec_by_insert_inherit_func(): On the update of PRIMARY KEY of a record that contained a dropped column whose value was stored externally, we will be inserting a dummy NULL or empty string value to the field of the dropped column. The externally stored column would eventually be dropped when purge removes the delete-marked record for the old PRIMARY KEY value. btr_index_rec_validate(): Recognize the metadata record. btr_discard_only_page_on_level(): Preserve the generic instant ALTER TABLE metadata. btr_set_instant(): Replaces page_set_instant(). This sets a clustered index root page to the appropriate format, or upgrades from the MDEV-11369 instant ADD COLUMN to generic ALTER TABLE format. btr_cur_instant_init_low(): Read and validate the metadata BLOB page before reconstructing the dictionary information based on it. btr_cur_instant_init_metadata(): Do not read any lengths from the metadata record header before reading the BLOB. At this point, we would not actually know how many nullable fields the metadata record contains. btr_cur_instant_root_init(): Initialize n_core_null_bytes in one of two possible ways. btr_cur_trim(): Handle the mblob record. row_metadata_to_tuple(): Convert a metadata record to a data tuple, based on the new info_bits of the metadata record. btr_cur_pessimistic_update(): Invoke row_metadata_to_tuple() if needed. Invoke dtuple_convert_big_rec() for metadata records if the record is too large, or if the mblob is not yet marked as externally stored. btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete(): When the last user record is deleted, do not delete the generic instant ALTER TABLE metadata record. Only delete MDEV-11369 instant ADD COLUMN metadata records. btr_cur_optimistic_insert(): Avoid unnecessary computation of rec_size. btr_pcur_store_position(): Allow a logically empty page to contain a metadata record for generic ALTER TABLE. REC_INFO_DEFAULT_ROW_ADD: Renamed from REC_INFO_DEFAULT_ROW. This is for the old instant ADD COLUMN (MDEV-11369) only. REC_INFO_DEFAULT_ROW_ALTER: The more generic metadata record, with additional information for dropped or reordered columns. rec_info_bits_valid(): Remove. The only case when this would fail is when the record is the generic ALTER TABLE metadata record. rec_is_alter_metadata(): Check if a record is the metadata record for instant ALTER TABLE (other than ADD COLUMN). NOTE: This function must not be invoked on node pointer records, because the delete-mark flag in those records may be set (it is garbage), and then a debug assertion could fail because index->is_instant() does not necessarily hold. rec_is_add_metadata(): Check if a record is MDEV-11369 ADD COLUMN metadata record (not more generic instant ALTER TABLE). rec_get_converted_size_comp_prefix_low(): Assume that the metadata field will be stored externally. In dtuple_convert_big_rec() during the rec_get_converted_size() call, it would not be there yet. rec_get_converted_size_comp(): Replace status,fields,n_fields with tuple. rec_init_offsets_comp_ordinary(), rec_get_converted_size_comp_prefix_low(), rec_convert_dtuple_to_rec_comp(): Add template<bool mblob = false>. With mblob=true, process a record with a metadata BLOB. rec_copy_prefix_to_buf(): Assert that no fields beyond the key and system columns are being copied. Exclude the metadata BLOB field. rec_convert_dtuple_to_metadata_comp(): Convert an alter metadata tuple into a record. row_upd_index_replace_metadata(): Apply an update vector to an alter_metadata tuple. row_log_allocate(): Replace dict_index_t::is_instant() with a more appropriate condition that ignores dict_table_t::instant. Only a table on which the MDEV-11369 ADD COLUMN was performed can "lose its instantness" when it becomes empty. After instant DROP COLUMN or reordering columns, we cannot simply convert the table to the canonical format, because the data dictionary cache and all possibly existing references to it from other client connection threads would have to be adjusted. row_quiesce_write_index_fields(): Do not crash when the table contains an instantly dropped column. Thanks to Thirunarayanan Balathandayuthapani for discussing the design and implementing an initial prototype of this. Thanks to Matthias Leich for testing.
7 years ago
11 years ago
11 years ago
MDEV-10139 Support for InnoDB SEQUENCE objects We introduce a NO_ROLLBACK flag for InnoDB tables. This flag only works for tables that have a single index. Apart from undo logging, this flag will also prevent locking and the assignment of DB_ROW_ID or DB_TRX_ID, and imply READ UNCOMMITTED isolation. It is assumed that the SQL layer is guaranteeing mutual exclusion. After the initial insert of the single record during CREATE SEQUENCE, InnoDB will be updating the single record in-place. This is crash-safe thanks to the redo log. (That is, after a crash after CREATE SEQUENCE was committed, the effect of sequence operations will be observable fully or not at all.) When it comes to the durability of the updates of SEQUENCE in InnoDB, there is a clear analogy to MDEV-6076 Persistent AUTO_INCREMENT. The updates would be made persistent by the InnoDB redo log flush at transaction commit or rollback (or XA PREPARE), provided that innodb_log_flush_at_trx_commit=1. Similar to AUTO_INCREMENT, it is possible that the update of a SEQUENCE in a middle of transaction becomes durable before the COMMIT/ROLLBACK of the transaction, in case the InnoDB redo log is being flushed as a result of the a commit or rollback of some other transaction, or as a result of a redo log checkpoint that can be initiated at any time by operations that are writing redo log. dict_table_t::no_rollback(): Check if the table does not support rollback. BTR_NO_ROLLBACK: Logging and locking flags for no_rollback() tables. DICT_TF_BITS: Add the NO_ROLLBACK flag. row_ins_step(): Assign 0 to DB_ROW_ID and DB_TRX_ID, and skip any locking for no-rollback tables. There will be only a single row in no-rollback tables (or there must be a proper PRIMARY KEY). row_search_mvcc(): Execute the READ UNCOMMITTED code path for no-rollback tables. ha_innobase::external_lock(), ha_innobase::store_lock(): Block CREATE/DROP SEQUENCE in innodb_read_only mode. This probably has no effect for CREATE SEQUENCE, because already ha_innobase::create() should have been called (and refused) before external_lock() or store_lock() is called. ha_innobase::store_lock(): For CREATE SEQUENCE, do not acquire any InnoDB locks, even though TL_WRITE is being requested. (This is just a performance optimization.) innobase_copy_frm_flags_from_create_info(), row_drop_table_for_mysql(): Disable persistent statistics for no_rollback tables.
9 years ago
Merge 10.1 into 10.2 This only merges MDEV-12253, adapting it to MDEV-12602 which is already present in 10.2 but not yet in the 10.1 revision that is being merged. TODO: Error handling in crash recovery needs to be improved. If a page cannot be decrypted (or read), we should cleanly abort the startup. If innodb_force_recovery is specified, we should ignore the problematic page and apply redo log to other pages. Currently, the test encryption.innodb-redo-badkey randomly fails like this (the last messages are from cmake -DWITH_ASAN): 2017-05-05 10:19:40 140037071685504 [Note] InnoDB: Starting crash recovery from checkpoint LSN=1635994 2017-05-05 10:19:40 140037071685504 [ERROR] InnoDB: Missing MLOG_FILE_NAME or MLOG_FILE_DELETE before MLOG_CHECKPOINT for tablespace 1 2017-05-05 10:19:40 140037071685504 [ERROR] InnoDB: Plugin initialization aborted at srv0start.cc[2201] with error Data structure corruption 2017-05-05 10:19:41 140037071685504 [Note] InnoDB: Starting shutdown... i================================================================= ==5226==ERROR: AddressSanitizer: attempting free on address which was not malloc()-ed: 0x612000018588 in thread T0 #0 0x736750 in operator delete(void*) (/mariadb/server/build/sql/mysqld+0x736750) #1 0x1e4833f in LatchCounter::~LatchCounter() /mariadb/server/storage/innobase/include/sync0types.h:599:4 #2 0x1e480b8 in LatchMeta<LatchCounter>::~LatchMeta() /mariadb/server/storage/innobase/include/sync0types.h:786:17 #3 0x1e35509 in sync_latch_meta_destroy() /mariadb/server/storage/innobase/sync/sync0debug.cc:1622:3 #4 0x1e35314 in sync_check_close() /mariadb/server/storage/innobase/sync/sync0debug.cc:1839:2 #5 0x1dfdc18 in innodb_shutdown() /mariadb/server/storage/innobase/srv/srv0start.cc:2888:2 #6 0x197e5e6 in innobase_init(void*) /mariadb/server/storage/innobase/handler/ha_innodb.cc:4475:3
9 years ago
MDEV-12266: Change dict_table_t::space to fil_space_t* InnoDB always keeps all tablespaces in the fil_system cache. The fil_system.LRU is only for closing file handles; the fil_space_t and fil_node_t for all data files will remain in main memory. Between startup to shutdown, they can only be created and removed by DDL statements. Therefore, we can let dict_table_t::space point directly to the fil_space_t. dict_table_t::space_id: A numeric tablespace ID for the corner cases where we do not have a tablespace. The most prominent examples are ALTER TABLE...DISCARD TABLESPACE or a missing or corrupted file. There are a few functional differences; most notably: (1) DROP TABLE will delete matching .ibd and .cfg files, even if they were not attached to the data dictionary. (2) Some error messages will report file names instead of numeric IDs. There still are many functions that use numeric tablespace IDs instead of fil_space_t*, and many functions could be converted to fil_space_t member functions. Also, Tablespace and Datafile should be merged with fil_space_t and fil_node_t. page_id_t and buf_page_get_gen() could use fil_space_t& instead of a numeric ID, and after moving to a single buffer pool (MDEV-15058), buf_pool_t::page_hash could be moved to fil_space_t::page_hash. FilSpace: Remove. Only few calls to fil_space_acquire() will remain, and gradually they should be removed. mtr_t::set_named_space_id(ulint): Renamed from set_named_space(), to prevent accidental calls to this slower function. Very few callers remain. fseg_create(), fsp_reserve_free_extents(): Take fil_space_t* as a parameter instead of a space_id. fil_space_t::rename(): Wrapper for fil_rename_tablespace_check(), fil_name_write_rename(), fil_rename_tablespace(). Mariabackup passes the parameter log=false; InnoDB passes log=true. dict_mem_table_create(): Take fil_space_t* instead of space_id as parameter. dict_process_sys_tables_rec_and_mtr_commit(): Replace the parameter 'status' with 'bool cached'. dict_get_and_save_data_dir_path(): Avoid copying the fil_node_t::name. fil_ibd_open(): Return the tablespace. fil_space_t::set_imported(): Replaces fil_space_set_imported(). truncate_t: Change many member function parameters to fil_space_t*, and remove page_size parameters. row_truncate_prepare(): Merge to its only caller. row_drop_table_from_cache(): Assert that the table is persistent. dict_create_sys_indexes_tuple(): Write SYS_INDEXES.SPACE=FIL_NULL if the tablespace has been discarded. row_import_update_discarded_flag(): Remove a constant parameter.
8 years ago
Merge 10.1 into 10.2 This only merges MDEV-12253, adapting it to MDEV-12602 which is already present in 10.2 but not yet in the 10.1 revision that is being merged. TODO: Error handling in crash recovery needs to be improved. If a page cannot be decrypted (or read), we should cleanly abort the startup. If innodb_force_recovery is specified, we should ignore the problematic page and apply redo log to other pages. Currently, the test encryption.innodb-redo-badkey randomly fails like this (the last messages are from cmake -DWITH_ASAN): 2017-05-05 10:19:40 140037071685504 [Note] InnoDB: Starting crash recovery from checkpoint LSN=1635994 2017-05-05 10:19:40 140037071685504 [ERROR] InnoDB: Missing MLOG_FILE_NAME or MLOG_FILE_DELETE before MLOG_CHECKPOINT for tablespace 1 2017-05-05 10:19:40 140037071685504 [ERROR] InnoDB: Plugin initialization aborted at srv0start.cc[2201] with error Data structure corruption 2017-05-05 10:19:41 140037071685504 [Note] InnoDB: Starting shutdown... i================================================================= ==5226==ERROR: AddressSanitizer: attempting free on address which was not malloc()-ed: 0x612000018588 in thread T0 #0 0x736750 in operator delete(void*) (/mariadb/server/build/sql/mysqld+0x736750) #1 0x1e4833f in LatchCounter::~LatchCounter() /mariadb/server/storage/innobase/include/sync0types.h:599:4 #2 0x1e480b8 in LatchMeta<LatchCounter>::~LatchMeta() /mariadb/server/storage/innobase/include/sync0types.h:786:17 #3 0x1e35509 in sync_latch_meta_destroy() /mariadb/server/storage/innobase/sync/sync0debug.cc:1622:3 #4 0x1e35314 in sync_check_close() /mariadb/server/storage/innobase/sync/sync0debug.cc:1839:2 #5 0x1dfdc18 in innodb_shutdown() /mariadb/server/storage/innobase/srv/srv0start.cc:2888:2 #6 0x197e5e6 in innobase_init(void*) /mariadb/server/storage/innobase/handler/ha_innodb.cc:4475:3
9 years ago
MDEV-10139 Support for InnoDB SEQUENCE objects We introduce a NO_ROLLBACK flag for InnoDB tables. This flag only works for tables that have a single index. Apart from undo logging, this flag will also prevent locking and the assignment of DB_ROW_ID or DB_TRX_ID, and imply READ UNCOMMITTED isolation. It is assumed that the SQL layer is guaranteeing mutual exclusion. After the initial insert of the single record during CREATE SEQUENCE, InnoDB will be updating the single record in-place. This is crash-safe thanks to the redo log. (That is, after a crash after CREATE SEQUENCE was committed, the effect of sequence operations will be observable fully or not at all.) When it comes to the durability of the updates of SEQUENCE in InnoDB, there is a clear analogy to MDEV-6076 Persistent AUTO_INCREMENT. The updates would be made persistent by the InnoDB redo log flush at transaction commit or rollback (or XA PREPARE), provided that innodb_log_flush_at_trx_commit=1. Similar to AUTO_INCREMENT, it is possible that the update of a SEQUENCE in a middle of transaction becomes durable before the COMMIT/ROLLBACK of the transaction, in case the InnoDB redo log is being flushed as a result of the a commit or rollback of some other transaction, or as a result of a redo log checkpoint that can be initiated at any time by operations that are writing redo log. dict_table_t::no_rollback(): Check if the table does not support rollback. BTR_NO_ROLLBACK: Logging and locking flags for no_rollback() tables. DICT_TF_BITS: Add the NO_ROLLBACK flag. row_ins_step(): Assign 0 to DB_ROW_ID and DB_TRX_ID, and skip any locking for no-rollback tables. There will be only a single row in no-rollback tables (or there must be a proper PRIMARY KEY). row_search_mvcc(): Execute the READ UNCOMMITTED code path for no-rollback tables. ha_innobase::external_lock(), ha_innobase::store_lock(): Block CREATE/DROP SEQUENCE in innodb_read_only mode. This probably has no effect for CREATE SEQUENCE, because already ha_innobase::create() should have been called (and refused) before external_lock() or store_lock() is called. ha_innobase::store_lock(): For CREATE SEQUENCE, do not acquire any InnoDB locks, even though TL_WRITE is being requested. (This is just a performance optimization.) innobase_copy_frm_flags_from_create_info(), row_drop_table_for_mysql(): Disable persistent statistics for no_rollback tables.
9 years ago
MDEV-15662 Instant DROP COLUMN or changing the order of columns Allow ADD COLUMN anywhere in a table, not only adding as the last column. Allow instant DROP COLUMN and instant changing the order of columns. The added columns will always be added last in clustered index records. In new records, instantly dropped columns will be stored as NULL or empty when possible. Information about dropped and reordered columns will be written in a metadata BLOB (mblob), which is stored before the first 'user' field in the hidden metadata record at the start of the clustered index. The presence of mblob is indicated by setting the delete-mark flag in the metadata record. The metadata BLOB stores the number of clustered index fields, followed by an array of column information for each field. For dropped columns, we store the NOT NULL flag, the fixed length, and for variable-length columns, whether the maximum length exceeded 255 bytes. For non-dropped columns, we store the column position. Unlike with MDEV-11369, when a table becomes empty, it cannot be converted back to the canonical format. The reason for this is that other threads may hold cached objects such as row_prebuilt_t::ins_node that could refer to dropped or reordered index fields. For instant DROP COLUMN and ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC, we must store the n_core_null_bytes in the root page, so that the chain of node pointer records can be followed in order to reach the leftmost leaf page where the metadata record is located. If the mblob is present, we will zero-initialize the strings "infimum" and "supremum" in the root page, and use the last byte of "supremum" for storing the number of null bytes (which are allocated but useless on node pointer pages). This is necessary for btr_cur_instant_init_metadata() to be able to navigate to the mblob. If the PRIMARY KEY contains any variable-length column and some nullable columns were instantly dropped, the dict_index_t::n_nullable in the data dictionary could be smaller than it actually is in the non-leaf pages. Because of this, the non-leaf pages could use more bytes for the null flags than the data dictionary expects, and we could be reading the lengths of the variable-length columns from the wrong offset, and thus reading the child page number from wrong place. This is the result of two design mistakes that involve unnecessary storage of data: First, it is nonsense to store any data fields for the leftmost node pointer records, because the comparisons would be resolved by the MIN_REC_FLAG alone. Second, there cannot be any null fields in the clustered index node pointer fields, but we nevertheless reserve space for all the null flags. Limitations (future work): MDEV-17459 Allow instant ALTER TABLE even if FULLTEXT INDEX exists MDEV-17468 Avoid table rebuild on operations on generated columns MDEV-17494 Refuse ALGORITHM=INSTANT when the row size is too large btr_page_reorganize_low(): Preserve any metadata in the root page. Call lock_move_reorganize_page() only after restoring the "infimum" and "supremum" records, to avoid a memcmp() assertion failure. dict_col_t::DROPPED: Magic value for dict_col_t::ind. dict_col_t::clear_instant(): Renamed from dict_col_t::remove_instant(). Do not assert that the column was instantly added, because we sometimes call this unconditionally for all columns. Convert an instantly added column to a "core column". The old name remove_instant() could be mistaken to refer to "instant DROP COLUMN". dict_col_t::is_added(): Rename from dict_col_t::is_instant(). dtype_t::metadata_blob_init(): Initialize the mblob data type. dtuple_t::is_metadata(), dtuple_t::is_alter_metadata(), upd_t::is_metadata(), upd_t::is_alter_metadata(): Check if info_bits refer to a metadata record. dict_table_t::instant: Metadata about dropped or reordered columns. dict_table_t::prepare_instant(): Prepare ha_innobase_inplace_ctx::instant_table for instant ALTER TABLE. innobase_instant_try() will pass this to dict_table_t::instant_column(). On rollback, dict_table_t::rollback_instant() will be called. dict_table_t::instant_column(): Renamed from instant_add_column(). Add the parameter col_map so that columns can be reordered. Copy and adjust v_cols[] as well. dict_table_t::find(): Find an old column based on a new column number. dict_table_t::serialise_columns(), dict_table_t::deserialise_columns(): Convert the mblob. dict_index_t::instant_metadata(): Create the metadata record for instant ALTER TABLE. Invoke dict_table_t::serialise_columns(). dict_index_t::reconstruct_fields(): Invoked by dict_table_t::deserialise_columns(). dict_index_t::clear_instant_alter(): Move the fields for the dropped columns to the end, and sort the surviving index fields in ascending order of column position. ha_innobase::check_if_supported_inplace_alter(): Do not allow adding a FTS_DOC_ID column if a hidden FTS_DOC_ID column exists due to FULLTEXT INDEX. (This always required ALGORITHM=COPY.) instant_alter_column_possible(): Add a parameter for InnoDB table, to check for additional conditions, such as the maximum number of index fields. ha_innobase_inplace_ctx::first_alter_pos: The first column whose position is affected by instant ADD, DROP, or changing the order of columns. innobase_build_col_map(): Skip added virtual columns. prepare_inplace_add_virtual(): Correctly compute num_to_add_vcol. Remove some unnecessary code. Note that the call to innodb_base_col_setup() should be executed later. commit_try_norebuild(): If ctx->is_instant(), let the virtual columns be added or dropped by innobase_instant_try(). innobase_instant_try(): Fill in a zero default value for the hidden column FTS_DOC_ID (to reduce the work needed in MDEV-17459). If any columns were dropped or reordered (or added not last), delete any SYS_COLUMNS records for the following columns, and insert SYS_COLUMNS records for all subsequent stored columns as well as for all virtual columns. If any virtual column is dropped, rewrite all virtual column metadata. Use a shortcut only for adding virtual columns. This is because innobase_drop_virtual_try() assumes that the dropped virtual columns still exist in ctx->old_table. innodb_update_cols(): Renamed from innodb_update_n_cols(). innobase_add_one_virtual(), innobase_insert_sys_virtual(): Change the return type to bool, and invoke my_error() when detecting an error. innodb_insert_sys_columns(): Insert a record into SYS_COLUMNS. Refactored from innobase_add_one_virtual() and innobase_instant_add_col(). innobase_instant_add_col(): Replace the parameter dfield with type. innobase_instant_drop_cols(): Drop matching columns from SYS_COLUMNS and all columns from SYS_VIRTUAL. innobase_add_virtual_try(), innobase_drop_virtual_try(): Let the caller invoke innodb_update_cols(). innobase_rename_column_try(): Skip dropped columns. commit_cache_norebuild(): Update table->fts->doc_col. dict_mem_table_col_rename_low(): Skip dropped columns. trx_undo_rec_get_partial_row(): Skip dropped columns. trx_undo_update_rec_get_update(): Handle the metadata BLOB correctly. trx_undo_page_report_modify(): Avoid out-of-bounds access to record fields. Log metadata records consistently. Apparently, the first fields of a clustered index may be updated in an update_undo vector when the index is ID_IND of SYS_FOREIGN, as part of renaming the table during ALTER TABLE. Normally, updates of the PRIMARY KEY should be logged as delete-mark and an insert. row_undo_mod_parse_undo_rec(), row_purge_parse_undo_rec(): Use trx_undo_metadata. row_undo_mod_clust_low(): On metadata rollback, roll back the root page too. row_undo_mod_clust(): Relax an assertion. The delete-mark flag was repurposed for ALTER TABLE metadata records. row_rec_to_index_entry_impl(): Add the template parameter mblob and the optional parameter info_bits for specifying the desired new info bits. For the metadata tuple, allow conversion between the original format (ADD COLUMN only) and the generic format (with hidden BLOB). Add the optional parameter "pad" to determine whether the tuple should be padded to the index fields (on ALTER TABLE it should), or whether it should remain at its original size (on rollback). row_build_index_entry_low(): Clean up the code, removing redundant variables and conditions. For instantly dropped columns, generate a dummy value that is NULL, the empty string, or a fixed length of NUL bytes, depending on the type of the dropped column. row_upd_clust_rec_by_insert_inherit_func(): On the update of PRIMARY KEY of a record that contained a dropped column whose value was stored externally, we will be inserting a dummy NULL or empty string value to the field of the dropped column. The externally stored column would eventually be dropped when purge removes the delete-marked record for the old PRIMARY KEY value. btr_index_rec_validate(): Recognize the metadata record. btr_discard_only_page_on_level(): Preserve the generic instant ALTER TABLE metadata. btr_set_instant(): Replaces page_set_instant(). This sets a clustered index root page to the appropriate format, or upgrades from the MDEV-11369 instant ADD COLUMN to generic ALTER TABLE format. btr_cur_instant_init_low(): Read and validate the metadata BLOB page before reconstructing the dictionary information based on it. btr_cur_instant_init_metadata(): Do not read any lengths from the metadata record header before reading the BLOB. At this point, we would not actually know how many nullable fields the metadata record contains. btr_cur_instant_root_init(): Initialize n_core_null_bytes in one of two possible ways. btr_cur_trim(): Handle the mblob record. row_metadata_to_tuple(): Convert a metadata record to a data tuple, based on the new info_bits of the metadata record. btr_cur_pessimistic_update(): Invoke row_metadata_to_tuple() if needed. Invoke dtuple_convert_big_rec() for metadata records if the record is too large, or if the mblob is not yet marked as externally stored. btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete(): When the last user record is deleted, do not delete the generic instant ALTER TABLE metadata record. Only delete MDEV-11369 instant ADD COLUMN metadata records. btr_cur_optimistic_insert(): Avoid unnecessary computation of rec_size. btr_pcur_store_position(): Allow a logically empty page to contain a metadata record for generic ALTER TABLE. REC_INFO_DEFAULT_ROW_ADD: Renamed from REC_INFO_DEFAULT_ROW. This is for the old instant ADD COLUMN (MDEV-11369) only. REC_INFO_DEFAULT_ROW_ALTER: The more generic metadata record, with additional information for dropped or reordered columns. rec_info_bits_valid(): Remove. The only case when this would fail is when the record is the generic ALTER TABLE metadata record. rec_is_alter_metadata(): Check if a record is the metadata record for instant ALTER TABLE (other than ADD COLUMN). NOTE: This function must not be invoked on node pointer records, because the delete-mark flag in those records may be set (it is garbage), and then a debug assertion could fail because index->is_instant() does not necessarily hold. rec_is_add_metadata(): Check if a record is MDEV-11369 ADD COLUMN metadata record (not more generic instant ALTER TABLE). rec_get_converted_size_comp_prefix_low(): Assume that the metadata field will be stored externally. In dtuple_convert_big_rec() during the rec_get_converted_size() call, it would not be there yet. rec_get_converted_size_comp(): Replace status,fields,n_fields with tuple. rec_init_offsets_comp_ordinary(), rec_get_converted_size_comp_prefix_low(), rec_convert_dtuple_to_rec_comp(): Add template<bool mblob = false>. With mblob=true, process a record with a metadata BLOB. rec_copy_prefix_to_buf(): Assert that no fields beyond the key and system columns are being copied. Exclude the metadata BLOB field. rec_convert_dtuple_to_metadata_comp(): Convert an alter metadata tuple into a record. row_upd_index_replace_metadata(): Apply an update vector to an alter_metadata tuple. row_log_allocate(): Replace dict_index_t::is_instant() with a more appropriate condition that ignores dict_table_t::instant. Only a table on which the MDEV-11369 ADD COLUMN was performed can "lose its instantness" when it becomes empty. After instant DROP COLUMN or reordering columns, we cannot simply convert the table to the canonical format, because the data dictionary cache and all possibly existing references to it from other client connection threads would have to be adjusted. row_quiesce_write_index_fields(): Do not crash when the table contains an instantly dropped column. Thanks to Thirunarayanan Balathandayuthapani for discussing the design and implementing an initial prototype of this. Thanks to Matthias Leich for testing.
7 years ago
MDEV-11369 Instant ADD COLUMN for InnoDB For InnoDB tables, adding, dropping and reordering columns has required a rebuild of the table and all its indexes. Since MySQL 5.6 (and MariaDB 10.0) this has been supported online (LOCK=NONE), allowing concurrent modification of the tables. This work revises the InnoDB ROW_FORMAT=REDUNDANT, ROW_FORMAT=COMPACT and ROW_FORMAT=DYNAMIC so that columns can be appended instantaneously, with only minor changes performed to the table structure. The counter innodb_instant_alter_column in INFORMATION_SCHEMA.GLOBAL_STATUS is incremented whenever a table rebuild operation is converted into an instant ADD COLUMN operation. ROW_FORMAT=COMPRESSED tables will not support instant ADD COLUMN. Some usability limitations will be addressed in subsequent work: MDEV-13134 Introduce ALTER TABLE attributes ALGORITHM=NOCOPY and ALGORITHM=INSTANT MDEV-14016 Allow instant ADD COLUMN, ADD INDEX, LOCK=NONE The format of the clustered index (PRIMARY KEY) is changed as follows: (1) The FIL_PAGE_TYPE of the root page will be FIL_PAGE_TYPE_INSTANT, and a new field PAGE_INSTANT will contain the original number of fields in the clustered index ('core' fields). If instant ADD COLUMN has not been used or the table becomes empty, or the very first instant ADD COLUMN operation is rolled back, the fields PAGE_INSTANT and FIL_PAGE_TYPE will be reset to 0 and FIL_PAGE_INDEX. (2) A special 'default row' record is inserted into the leftmost leaf, between the page infimum and the first user record. This record is distinguished by the REC_INFO_MIN_REC_FLAG, and it is otherwise in the same format as records that contain values for the instantly added columns. This 'default row' always has the same number of fields as the clustered index according to the table definition. The values of 'core' fields are to be ignored. For other fields, the 'default row' will contain the default values as they were during the ALTER TABLE statement. (If the column default values are changed later, those values will only be stored in the .frm file. The 'default row' will contain the original evaluated values, which must be the same for every row.) The 'default row' must be completely hidden from higher-level access routines. Assertions have been added to ensure that no 'default row' is ever present in the adaptive hash index or in locked records. The 'default row' is never delete-marked. (3) In clustered index leaf page records, the number of fields must reside between the number of 'core' fields (dict_index_t::n_core_fields introduced in this work) and dict_index_t::n_fields. If the number of fields is less than dict_index_t::n_fields, the missing fields are replaced with the column value of the 'default row'. Note: The number of fields in the record may shrink if some of the last instantly added columns are updated to the value that is in the 'default row'. The function btr_cur_trim() implements this 'compression' on update and rollback; dtuple::trim() implements it on insert. (4) In ROW_FORMAT=COMPACT and ROW_FORMAT=DYNAMIC records, the new status value REC_STATUS_COLUMNS_ADDED will indicate the presence of a new record header that will encode n_fields-n_core_fields-1 in 1 or 2 bytes. (In ROW_FORMAT=REDUNDANT records, the record header always explicitly encodes the number of fields.) We introduce the undo log record type TRX_UNDO_INSERT_DEFAULT for covering the insert of the 'default row' record when instant ADD COLUMN is used for the first time. Subsequent instant ADD COLUMN can use TRX_UNDO_UPD_EXIST_REC. This is joint work with Vin Chen (陈福荣) from Tencent. The design that was discussed in April 2017 would not have allowed import or export of data files, because instead of the 'default row' it would have introduced a data dictionary table. The test rpl.rpl_alter_instant is exactly as contributed in pull request #408. The test innodb.instant_alter is based on a contributed test. The redo log record format changes for ROW_FORMAT=DYNAMIC and ROW_FORMAT=COMPACT are as contributed. (With this change present, crash recovery from MariaDB 10.3.1 will fail in spectacular ways!) Also the semantics of higher-level redo log records that modify the PAGE_INSTANT field is changed. The redo log format version identifier was already changed to LOG_HEADER_FORMAT_CURRENT=103 in MariaDB 10.3.1. Everything else has been rewritten by me. Thanks to Elena Stepanova, the code has been tested extensively. When rolling back an instant ADD COLUMN operation, we must empty the PAGE_FREE list after deleting or shortening the 'default row' record, by calling either btr_page_empty() or btr_page_reorganize(). We must know the size of each entry in the PAGE_FREE list. If rollback left a freed copy of the 'default row' in the PAGE_FREE list, we would be unable to determine its size (if it is in ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC) because it would contain more fields than the rolled-back definition of the clustered index. UNIV_SQL_DEFAULT: A new special constant that designates an instantly added column that is not present in the clustered index record. len_is_stored(): Check if a length is an actual length. There are two magic length values: UNIV_SQL_DEFAULT, UNIV_SQL_NULL. dict_col_t::def_val: The 'default row' value of the column. If the column is not added instantly, def_val.len will be UNIV_SQL_DEFAULT. dict_col_t: Add the accessors is_virtual(), is_nullable(), is_instant(), instant_value(). dict_col_t::remove_instant(): Remove the 'instant ADD' status of a column. dict_col_t::name(const dict_table_t& table): Replaces dict_table_get_col_name(). dict_index_t::n_core_fields: The original number of fields. For secondary indexes and if instant ADD COLUMN has not been used, this will be equal to dict_index_t::n_fields. dict_index_t::n_core_null_bytes: Number of bytes needed to represent the null flags; usually equal to UT_BITS_IN_BYTES(n_nullable). dict_index_t::NO_CORE_NULL_BYTES: Magic value signalling that n_core_null_bytes was not initialized yet from the clustered index root page. dict_index_t: Add the accessors is_instant(), is_clust(), get_n_nullable(), instant_field_value(). dict_index_t::instant_add_field(): Adjust clustered index metadata for instant ADD COLUMN. dict_index_t::remove_instant(): Remove the 'instant ADD' status of a clustered index when the table becomes empty, or the very first instant ADD COLUMN operation is rolled back. dict_table_t: Add the accessors is_instant(), is_temporary(), supports_instant(). dict_table_t::instant_add_column(): Adjust metadata for instant ADD COLUMN. dict_table_t::rollback_instant(): Adjust metadata on the rollback of instant ADD COLUMN. prepare_inplace_alter_table_dict(): First create the ctx->new_table, and only then decide if the table really needs to be rebuilt. We must split the creation of table or index metadata from the creation of the dictionary table records and the creation of the data. In this way, we can transform a table-rebuilding operation into an instant ADD COLUMN operation. Dictionary objects will only be added to cache when table rebuilding or index creation is needed. The ctx->instant_table will never be added to cache. dict_table_t::add_to_cache(): Modified and renamed from dict_table_add_to_cache(). Do not modify the table metadata. Let the callers invoke dict_table_add_system_columns() and if needed, set can_be_evicted. dict_create_sys_tables_tuple(), dict_create_table_step(): Omit the system columns (which will now exist in the dict_table_t object already at this point). dict_create_table_step(): Expect the callers to invoke dict_table_add_system_columns(). pars_create_table(): Before creating the table creation execution graph, invoke dict_table_add_system_columns(). row_create_table_for_mysql(): Expect all callers to invoke dict_table_add_system_columns(). create_index_dict(): Replaces row_merge_create_index_graph(). innodb_update_n_cols(): Renamed from innobase_update_n_virtual(). Call my_error() if an error occurs. btr_cur_instant_init(), btr_cur_instant_init_low(), btr_cur_instant_root_init(): Load additional metadata from the clustered index and set dict_index_t::n_core_null_bytes. This is invoked when table metadata is first loaded into the data dictionary. dict_boot(): Initialize n_core_null_bytes for the four hard-coded dictionary tables. dict_create_index_step(): Initialize n_core_null_bytes. This is executed as part of CREATE TABLE. dict_index_build_internal_clust(): Initialize n_core_null_bytes to NO_CORE_NULL_BYTES if table->supports_instant(). row_create_index_for_mysql(): Initialize n_core_null_bytes for CREATE TEMPORARY TABLE. commit_cache_norebuild(): Call the code to rename or enlarge columns in the cache only if instant ADD COLUMN is not being used. (Instant ADD COLUMN would copy all column metadata from instant_table to old_table, including the names and lengths.) PAGE_INSTANT: A new 13-bit field for storing dict_index_t::n_core_fields. This is repurposing the 16-bit field PAGE_DIRECTION, of which only the least significant 3 bits were used. The original byte containing PAGE_DIRECTION will be accessible via the new constant PAGE_DIRECTION_B. page_get_instant(), page_set_instant(): Accessors for the PAGE_INSTANT. page_ptr_get_direction(), page_get_direction(), page_ptr_set_direction(): Accessors for PAGE_DIRECTION. page_direction_reset(): Reset PAGE_DIRECTION, PAGE_N_DIRECTION. page_direction_increment(): Increment PAGE_N_DIRECTION and set PAGE_DIRECTION. rec_get_offsets(): Use the 'leaf' parameter for non-debug purposes, and assume that heap_no is always set. Initialize all dict_index_t::n_fields for ROW_FORMAT=REDUNDANT records, even if the record contains fewer fields. rec_offs_make_valid(): Add the parameter 'leaf'. rec_copy_prefix_to_dtuple(): Assert that the tuple is only built on the core fields. Instant ADD COLUMN only applies to the clustered index, and we should never build a search key that has more than the PRIMARY KEY and possibly DB_TRX_ID,DB_ROLL_PTR. All these columns are always present. dict_index_build_data_tuple(): Remove assertions that would be duplicated in rec_copy_prefix_to_dtuple(). rec_init_offsets(): Support ROW_FORMAT=REDUNDANT records whose number of fields is between n_core_fields and n_fields. cmp_rec_rec_with_match(): Implement the comparison between two MIN_REC_FLAG records. trx_t::in_rollback: Make the field available in non-debug builds. trx_start_for_ddl_low(): Remove dangerous error-tolerance. A dictionary transaction must be flagged as such before it has generated any undo log records. This is because trx_undo_assign_undo() will mark the transaction as a dictionary transaction in the undo log header right before the very first undo log record is being written. btr_index_rec_validate(): Account for instant ADD COLUMN row_undo_ins_remove_clust_rec(): On the rollback of an insert into SYS_COLUMNS, revert instant ADD COLUMN in the cache by removing the last column from the table and the clustered index. row_search_on_row_ref(), row_undo_mod_parse_undo_rec(), row_undo_mod(), trx_undo_update_rec_get_update(): Handle the 'default row' as a special case. dtuple_t::trim(index): Omit a redundant suffix of an index tuple right before insert or update. After instant ADD COLUMN, if the last fields of a clustered index tuple match the 'default row', there is no need to store them. While trimming the entry, we must hold a page latch, so that the table cannot be emptied and the 'default row' be deleted. btr_cur_optimistic_update(), btr_cur_pessimistic_update(), row_upd_clust_rec_by_insert(), row_ins_clust_index_entry_low(): Invoke dtuple_t::trim() if needed. row_ins_clust_index_entry(): Restore dtuple_t::n_fields after calling row_ins_clust_index_entry_low(). rec_get_converted_size(), rec_get_converted_size_comp(): Allow the number of fields to be between n_core_fields and n_fields. Do not support infimum,supremum. They are never supposed to be stored in dtuple_t, because page creation nowadays uses a lower-level method for initializing them. rec_convert_dtuple_to_rec_comp(): Assign the status bits based on the number of fields. btr_cur_trim(): In an update, trim the index entry as needed. For the 'default row', handle rollback specially. For user records, omit fields that match the 'default row'. btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete(): Skip locking and adaptive hash index for the 'default row'. row_log_table_apply_convert_mrec(): Replace 'default row' values if needed. In the temporary file that is applied by row_log_table_apply(), we must identify whether the records contain the extra header for instantly added columns. For now, we will allocate an additional byte for this for ROW_T_INSERT and ROW_T_UPDATE records when the source table has been subject to instant ADD COLUMN. The ROW_T_DELETE records are fine, as they will be converted and will only contain 'core' columns (PRIMARY KEY and some system columns) that are converted from dtuple_t. rec_get_converted_size_temp(), rec_init_offsets_temp(), rec_convert_dtuple_to_temp(): Add the parameter 'status'. REC_INFO_DEFAULT_ROW = REC_INFO_MIN_REC_FLAG | REC_STATUS_COLUMNS_ADDED: An info_bits constant for distinguishing the 'default row' record. rec_comp_status_t: An enum of the status bit values. rec_leaf_format: An enum that replaces the bool parameter of rec_init_offsets_comp_ordinary().
8 years ago
MDEV-15662 Instant DROP COLUMN or changing the order of columns Allow ADD COLUMN anywhere in a table, not only adding as the last column. Allow instant DROP COLUMN and instant changing the order of columns. The added columns will always be added last in clustered index records. In new records, instantly dropped columns will be stored as NULL or empty when possible. Information about dropped and reordered columns will be written in a metadata BLOB (mblob), which is stored before the first 'user' field in the hidden metadata record at the start of the clustered index. The presence of mblob is indicated by setting the delete-mark flag in the metadata record. The metadata BLOB stores the number of clustered index fields, followed by an array of column information for each field. For dropped columns, we store the NOT NULL flag, the fixed length, and for variable-length columns, whether the maximum length exceeded 255 bytes. For non-dropped columns, we store the column position. Unlike with MDEV-11369, when a table becomes empty, it cannot be converted back to the canonical format. The reason for this is that other threads may hold cached objects such as row_prebuilt_t::ins_node that could refer to dropped or reordered index fields. For instant DROP COLUMN and ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC, we must store the n_core_null_bytes in the root page, so that the chain of node pointer records can be followed in order to reach the leftmost leaf page where the metadata record is located. If the mblob is present, we will zero-initialize the strings "infimum" and "supremum" in the root page, and use the last byte of "supremum" for storing the number of null bytes (which are allocated but useless on node pointer pages). This is necessary for btr_cur_instant_init_metadata() to be able to navigate to the mblob. If the PRIMARY KEY contains any variable-length column and some nullable columns were instantly dropped, the dict_index_t::n_nullable in the data dictionary could be smaller than it actually is in the non-leaf pages. Because of this, the non-leaf pages could use more bytes for the null flags than the data dictionary expects, and we could be reading the lengths of the variable-length columns from the wrong offset, and thus reading the child page number from wrong place. This is the result of two design mistakes that involve unnecessary storage of data: First, it is nonsense to store any data fields for the leftmost node pointer records, because the comparisons would be resolved by the MIN_REC_FLAG alone. Second, there cannot be any null fields in the clustered index node pointer fields, but we nevertheless reserve space for all the null flags. Limitations (future work): MDEV-17459 Allow instant ALTER TABLE even if FULLTEXT INDEX exists MDEV-17468 Avoid table rebuild on operations on generated columns MDEV-17494 Refuse ALGORITHM=INSTANT when the row size is too large btr_page_reorganize_low(): Preserve any metadata in the root page. Call lock_move_reorganize_page() only after restoring the "infimum" and "supremum" records, to avoid a memcmp() assertion failure. dict_col_t::DROPPED: Magic value for dict_col_t::ind. dict_col_t::clear_instant(): Renamed from dict_col_t::remove_instant(). Do not assert that the column was instantly added, because we sometimes call this unconditionally for all columns. Convert an instantly added column to a "core column". The old name remove_instant() could be mistaken to refer to "instant DROP COLUMN". dict_col_t::is_added(): Rename from dict_col_t::is_instant(). dtype_t::metadata_blob_init(): Initialize the mblob data type. dtuple_t::is_metadata(), dtuple_t::is_alter_metadata(), upd_t::is_metadata(), upd_t::is_alter_metadata(): Check if info_bits refer to a metadata record. dict_table_t::instant: Metadata about dropped or reordered columns. dict_table_t::prepare_instant(): Prepare ha_innobase_inplace_ctx::instant_table for instant ALTER TABLE. innobase_instant_try() will pass this to dict_table_t::instant_column(). On rollback, dict_table_t::rollback_instant() will be called. dict_table_t::instant_column(): Renamed from instant_add_column(). Add the parameter col_map so that columns can be reordered. Copy and adjust v_cols[] as well. dict_table_t::find(): Find an old column based on a new column number. dict_table_t::serialise_columns(), dict_table_t::deserialise_columns(): Convert the mblob. dict_index_t::instant_metadata(): Create the metadata record for instant ALTER TABLE. Invoke dict_table_t::serialise_columns(). dict_index_t::reconstruct_fields(): Invoked by dict_table_t::deserialise_columns(). dict_index_t::clear_instant_alter(): Move the fields for the dropped columns to the end, and sort the surviving index fields in ascending order of column position. ha_innobase::check_if_supported_inplace_alter(): Do not allow adding a FTS_DOC_ID column if a hidden FTS_DOC_ID column exists due to FULLTEXT INDEX. (This always required ALGORITHM=COPY.) instant_alter_column_possible(): Add a parameter for InnoDB table, to check for additional conditions, such as the maximum number of index fields. ha_innobase_inplace_ctx::first_alter_pos: The first column whose position is affected by instant ADD, DROP, or changing the order of columns. innobase_build_col_map(): Skip added virtual columns. prepare_inplace_add_virtual(): Correctly compute num_to_add_vcol. Remove some unnecessary code. Note that the call to innodb_base_col_setup() should be executed later. commit_try_norebuild(): If ctx->is_instant(), let the virtual columns be added or dropped by innobase_instant_try(). innobase_instant_try(): Fill in a zero default value for the hidden column FTS_DOC_ID (to reduce the work needed in MDEV-17459). If any columns were dropped or reordered (or added not last), delete any SYS_COLUMNS records for the following columns, and insert SYS_COLUMNS records for all subsequent stored columns as well as for all virtual columns. If any virtual column is dropped, rewrite all virtual column metadata. Use a shortcut only for adding virtual columns. This is because innobase_drop_virtual_try() assumes that the dropped virtual columns still exist in ctx->old_table. innodb_update_cols(): Renamed from innodb_update_n_cols(). innobase_add_one_virtual(), innobase_insert_sys_virtual(): Change the return type to bool, and invoke my_error() when detecting an error. innodb_insert_sys_columns(): Insert a record into SYS_COLUMNS. Refactored from innobase_add_one_virtual() and innobase_instant_add_col(). innobase_instant_add_col(): Replace the parameter dfield with type. innobase_instant_drop_cols(): Drop matching columns from SYS_COLUMNS and all columns from SYS_VIRTUAL. innobase_add_virtual_try(), innobase_drop_virtual_try(): Let the caller invoke innodb_update_cols(). innobase_rename_column_try(): Skip dropped columns. commit_cache_norebuild(): Update table->fts->doc_col. dict_mem_table_col_rename_low(): Skip dropped columns. trx_undo_rec_get_partial_row(): Skip dropped columns. trx_undo_update_rec_get_update(): Handle the metadata BLOB correctly. trx_undo_page_report_modify(): Avoid out-of-bounds access to record fields. Log metadata records consistently. Apparently, the first fields of a clustered index may be updated in an update_undo vector when the index is ID_IND of SYS_FOREIGN, as part of renaming the table during ALTER TABLE. Normally, updates of the PRIMARY KEY should be logged as delete-mark and an insert. row_undo_mod_parse_undo_rec(), row_purge_parse_undo_rec(): Use trx_undo_metadata. row_undo_mod_clust_low(): On metadata rollback, roll back the root page too. row_undo_mod_clust(): Relax an assertion. The delete-mark flag was repurposed for ALTER TABLE metadata records. row_rec_to_index_entry_impl(): Add the template parameter mblob and the optional parameter info_bits for specifying the desired new info bits. For the metadata tuple, allow conversion between the original format (ADD COLUMN only) and the generic format (with hidden BLOB). Add the optional parameter "pad" to determine whether the tuple should be padded to the index fields (on ALTER TABLE it should), or whether it should remain at its original size (on rollback). row_build_index_entry_low(): Clean up the code, removing redundant variables and conditions. For instantly dropped columns, generate a dummy value that is NULL, the empty string, or a fixed length of NUL bytes, depending on the type of the dropped column. row_upd_clust_rec_by_insert_inherit_func(): On the update of PRIMARY KEY of a record that contained a dropped column whose value was stored externally, we will be inserting a dummy NULL or empty string value to the field of the dropped column. The externally stored column would eventually be dropped when purge removes the delete-marked record for the old PRIMARY KEY value. btr_index_rec_validate(): Recognize the metadata record. btr_discard_only_page_on_level(): Preserve the generic instant ALTER TABLE metadata. btr_set_instant(): Replaces page_set_instant(). This sets a clustered index root page to the appropriate format, or upgrades from the MDEV-11369 instant ADD COLUMN to generic ALTER TABLE format. btr_cur_instant_init_low(): Read and validate the metadata BLOB page before reconstructing the dictionary information based on it. btr_cur_instant_init_metadata(): Do not read any lengths from the metadata record header before reading the BLOB. At this point, we would not actually know how many nullable fields the metadata record contains. btr_cur_instant_root_init(): Initialize n_core_null_bytes in one of two possible ways. btr_cur_trim(): Handle the mblob record. row_metadata_to_tuple(): Convert a metadata record to a data tuple, based on the new info_bits of the metadata record. btr_cur_pessimistic_update(): Invoke row_metadata_to_tuple() if needed. Invoke dtuple_convert_big_rec() for metadata records if the record is too large, or if the mblob is not yet marked as externally stored. btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete(): When the last user record is deleted, do not delete the generic instant ALTER TABLE metadata record. Only delete MDEV-11369 instant ADD COLUMN metadata records. btr_cur_optimistic_insert(): Avoid unnecessary computation of rec_size. btr_pcur_store_position(): Allow a logically empty page to contain a metadata record for generic ALTER TABLE. REC_INFO_DEFAULT_ROW_ADD: Renamed from REC_INFO_DEFAULT_ROW. This is for the old instant ADD COLUMN (MDEV-11369) only. REC_INFO_DEFAULT_ROW_ALTER: The more generic metadata record, with additional information for dropped or reordered columns. rec_info_bits_valid(): Remove. The only case when this would fail is when the record is the generic ALTER TABLE metadata record. rec_is_alter_metadata(): Check if a record is the metadata record for instant ALTER TABLE (other than ADD COLUMN). NOTE: This function must not be invoked on node pointer records, because the delete-mark flag in those records may be set (it is garbage), and then a debug assertion could fail because index->is_instant() does not necessarily hold. rec_is_add_metadata(): Check if a record is MDEV-11369 ADD COLUMN metadata record (not more generic instant ALTER TABLE). rec_get_converted_size_comp_prefix_low(): Assume that the metadata field will be stored externally. In dtuple_convert_big_rec() during the rec_get_converted_size() call, it would not be there yet. rec_get_converted_size_comp(): Replace status,fields,n_fields with tuple. rec_init_offsets_comp_ordinary(), rec_get_converted_size_comp_prefix_low(), rec_convert_dtuple_to_rec_comp(): Add template<bool mblob = false>. With mblob=true, process a record with a metadata BLOB. rec_copy_prefix_to_buf(): Assert that no fields beyond the key and system columns are being copied. Exclude the metadata BLOB field. rec_convert_dtuple_to_metadata_comp(): Convert an alter metadata tuple into a record. row_upd_index_replace_metadata(): Apply an update vector to an alter_metadata tuple. row_log_allocate(): Replace dict_index_t::is_instant() with a more appropriate condition that ignores dict_table_t::instant. Only a table on which the MDEV-11369 ADD COLUMN was performed can "lose its instantness" when it becomes empty. After instant DROP COLUMN or reordering columns, we cannot simply convert the table to the canonical format, because the data dictionary cache and all possibly existing references to it from other client connection threads would have to be adjusted. row_quiesce_write_index_fields(): Do not crash when the table contains an instantly dropped column. Thanks to Thirunarayanan Balathandayuthapani for discussing the design and implementing an initial prototype of this. Thanks to Matthias Leich for testing.
7 years ago
MDEV-11369 Instant ADD COLUMN for InnoDB For InnoDB tables, adding, dropping and reordering columns has required a rebuild of the table and all its indexes. Since MySQL 5.6 (and MariaDB 10.0) this has been supported online (LOCK=NONE), allowing concurrent modification of the tables. This work revises the InnoDB ROW_FORMAT=REDUNDANT, ROW_FORMAT=COMPACT and ROW_FORMAT=DYNAMIC so that columns can be appended instantaneously, with only minor changes performed to the table structure. The counter innodb_instant_alter_column in INFORMATION_SCHEMA.GLOBAL_STATUS is incremented whenever a table rebuild operation is converted into an instant ADD COLUMN operation. ROW_FORMAT=COMPRESSED tables will not support instant ADD COLUMN. Some usability limitations will be addressed in subsequent work: MDEV-13134 Introduce ALTER TABLE attributes ALGORITHM=NOCOPY and ALGORITHM=INSTANT MDEV-14016 Allow instant ADD COLUMN, ADD INDEX, LOCK=NONE The format of the clustered index (PRIMARY KEY) is changed as follows: (1) The FIL_PAGE_TYPE of the root page will be FIL_PAGE_TYPE_INSTANT, and a new field PAGE_INSTANT will contain the original number of fields in the clustered index ('core' fields). If instant ADD COLUMN has not been used or the table becomes empty, or the very first instant ADD COLUMN operation is rolled back, the fields PAGE_INSTANT and FIL_PAGE_TYPE will be reset to 0 and FIL_PAGE_INDEX. (2) A special 'default row' record is inserted into the leftmost leaf, between the page infimum and the first user record. This record is distinguished by the REC_INFO_MIN_REC_FLAG, and it is otherwise in the same format as records that contain values for the instantly added columns. This 'default row' always has the same number of fields as the clustered index according to the table definition. The values of 'core' fields are to be ignored. For other fields, the 'default row' will contain the default values as they were during the ALTER TABLE statement. (If the column default values are changed later, those values will only be stored in the .frm file. The 'default row' will contain the original evaluated values, which must be the same for every row.) The 'default row' must be completely hidden from higher-level access routines. Assertions have been added to ensure that no 'default row' is ever present in the adaptive hash index or in locked records. The 'default row' is never delete-marked. (3) In clustered index leaf page records, the number of fields must reside between the number of 'core' fields (dict_index_t::n_core_fields introduced in this work) and dict_index_t::n_fields. If the number of fields is less than dict_index_t::n_fields, the missing fields are replaced with the column value of the 'default row'. Note: The number of fields in the record may shrink if some of the last instantly added columns are updated to the value that is in the 'default row'. The function btr_cur_trim() implements this 'compression' on update and rollback; dtuple::trim() implements it on insert. (4) In ROW_FORMAT=COMPACT and ROW_FORMAT=DYNAMIC records, the new status value REC_STATUS_COLUMNS_ADDED will indicate the presence of a new record header that will encode n_fields-n_core_fields-1 in 1 or 2 bytes. (In ROW_FORMAT=REDUNDANT records, the record header always explicitly encodes the number of fields.) We introduce the undo log record type TRX_UNDO_INSERT_DEFAULT for covering the insert of the 'default row' record when instant ADD COLUMN is used for the first time. Subsequent instant ADD COLUMN can use TRX_UNDO_UPD_EXIST_REC. This is joint work with Vin Chen (陈福荣) from Tencent. The design that was discussed in April 2017 would not have allowed import or export of data files, because instead of the 'default row' it would have introduced a data dictionary table. The test rpl.rpl_alter_instant is exactly as contributed in pull request #408. The test innodb.instant_alter is based on a contributed test. The redo log record format changes for ROW_FORMAT=DYNAMIC and ROW_FORMAT=COMPACT are as contributed. (With this change present, crash recovery from MariaDB 10.3.1 will fail in spectacular ways!) Also the semantics of higher-level redo log records that modify the PAGE_INSTANT field is changed. The redo log format version identifier was already changed to LOG_HEADER_FORMAT_CURRENT=103 in MariaDB 10.3.1. Everything else has been rewritten by me. Thanks to Elena Stepanova, the code has been tested extensively. When rolling back an instant ADD COLUMN operation, we must empty the PAGE_FREE list after deleting or shortening the 'default row' record, by calling either btr_page_empty() or btr_page_reorganize(). We must know the size of each entry in the PAGE_FREE list. If rollback left a freed copy of the 'default row' in the PAGE_FREE list, we would be unable to determine its size (if it is in ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC) because it would contain more fields than the rolled-back definition of the clustered index. UNIV_SQL_DEFAULT: A new special constant that designates an instantly added column that is not present in the clustered index record. len_is_stored(): Check if a length is an actual length. There are two magic length values: UNIV_SQL_DEFAULT, UNIV_SQL_NULL. dict_col_t::def_val: The 'default row' value of the column. If the column is not added instantly, def_val.len will be UNIV_SQL_DEFAULT. dict_col_t: Add the accessors is_virtual(), is_nullable(), is_instant(), instant_value(). dict_col_t::remove_instant(): Remove the 'instant ADD' status of a column. dict_col_t::name(const dict_table_t& table): Replaces dict_table_get_col_name(). dict_index_t::n_core_fields: The original number of fields. For secondary indexes and if instant ADD COLUMN has not been used, this will be equal to dict_index_t::n_fields. dict_index_t::n_core_null_bytes: Number of bytes needed to represent the null flags; usually equal to UT_BITS_IN_BYTES(n_nullable). dict_index_t::NO_CORE_NULL_BYTES: Magic value signalling that n_core_null_bytes was not initialized yet from the clustered index root page. dict_index_t: Add the accessors is_instant(), is_clust(), get_n_nullable(), instant_field_value(). dict_index_t::instant_add_field(): Adjust clustered index metadata for instant ADD COLUMN. dict_index_t::remove_instant(): Remove the 'instant ADD' status of a clustered index when the table becomes empty, or the very first instant ADD COLUMN operation is rolled back. dict_table_t: Add the accessors is_instant(), is_temporary(), supports_instant(). dict_table_t::instant_add_column(): Adjust metadata for instant ADD COLUMN. dict_table_t::rollback_instant(): Adjust metadata on the rollback of instant ADD COLUMN. prepare_inplace_alter_table_dict(): First create the ctx->new_table, and only then decide if the table really needs to be rebuilt. We must split the creation of table or index metadata from the creation of the dictionary table records and the creation of the data. In this way, we can transform a table-rebuilding operation into an instant ADD COLUMN operation. Dictionary objects will only be added to cache when table rebuilding or index creation is needed. The ctx->instant_table will never be added to cache. dict_table_t::add_to_cache(): Modified and renamed from dict_table_add_to_cache(). Do not modify the table metadata. Let the callers invoke dict_table_add_system_columns() and if needed, set can_be_evicted. dict_create_sys_tables_tuple(), dict_create_table_step(): Omit the system columns (which will now exist in the dict_table_t object already at this point). dict_create_table_step(): Expect the callers to invoke dict_table_add_system_columns(). pars_create_table(): Before creating the table creation execution graph, invoke dict_table_add_system_columns(). row_create_table_for_mysql(): Expect all callers to invoke dict_table_add_system_columns(). create_index_dict(): Replaces row_merge_create_index_graph(). innodb_update_n_cols(): Renamed from innobase_update_n_virtual(). Call my_error() if an error occurs. btr_cur_instant_init(), btr_cur_instant_init_low(), btr_cur_instant_root_init(): Load additional metadata from the clustered index and set dict_index_t::n_core_null_bytes. This is invoked when table metadata is first loaded into the data dictionary. dict_boot(): Initialize n_core_null_bytes for the four hard-coded dictionary tables. dict_create_index_step(): Initialize n_core_null_bytes. This is executed as part of CREATE TABLE. dict_index_build_internal_clust(): Initialize n_core_null_bytes to NO_CORE_NULL_BYTES if table->supports_instant(). row_create_index_for_mysql(): Initialize n_core_null_bytes for CREATE TEMPORARY TABLE. commit_cache_norebuild(): Call the code to rename or enlarge columns in the cache only if instant ADD COLUMN is not being used. (Instant ADD COLUMN would copy all column metadata from instant_table to old_table, including the names and lengths.) PAGE_INSTANT: A new 13-bit field for storing dict_index_t::n_core_fields. This is repurposing the 16-bit field PAGE_DIRECTION, of which only the least significant 3 bits were used. The original byte containing PAGE_DIRECTION will be accessible via the new constant PAGE_DIRECTION_B. page_get_instant(), page_set_instant(): Accessors for the PAGE_INSTANT. page_ptr_get_direction(), page_get_direction(), page_ptr_set_direction(): Accessors for PAGE_DIRECTION. page_direction_reset(): Reset PAGE_DIRECTION, PAGE_N_DIRECTION. page_direction_increment(): Increment PAGE_N_DIRECTION and set PAGE_DIRECTION. rec_get_offsets(): Use the 'leaf' parameter for non-debug purposes, and assume that heap_no is always set. Initialize all dict_index_t::n_fields for ROW_FORMAT=REDUNDANT records, even if the record contains fewer fields. rec_offs_make_valid(): Add the parameter 'leaf'. rec_copy_prefix_to_dtuple(): Assert that the tuple is only built on the core fields. Instant ADD COLUMN only applies to the clustered index, and we should never build a search key that has more than the PRIMARY KEY and possibly DB_TRX_ID,DB_ROLL_PTR. All these columns are always present. dict_index_build_data_tuple(): Remove assertions that would be duplicated in rec_copy_prefix_to_dtuple(). rec_init_offsets(): Support ROW_FORMAT=REDUNDANT records whose number of fields is between n_core_fields and n_fields. cmp_rec_rec_with_match(): Implement the comparison between two MIN_REC_FLAG records. trx_t::in_rollback: Make the field available in non-debug builds. trx_start_for_ddl_low(): Remove dangerous error-tolerance. A dictionary transaction must be flagged as such before it has generated any undo log records. This is because trx_undo_assign_undo() will mark the transaction as a dictionary transaction in the undo log header right before the very first undo log record is being written. btr_index_rec_validate(): Account for instant ADD COLUMN row_undo_ins_remove_clust_rec(): On the rollback of an insert into SYS_COLUMNS, revert instant ADD COLUMN in the cache by removing the last column from the table and the clustered index. row_search_on_row_ref(), row_undo_mod_parse_undo_rec(), row_undo_mod(), trx_undo_update_rec_get_update(): Handle the 'default row' as a special case. dtuple_t::trim(index): Omit a redundant suffix of an index tuple right before insert or update. After instant ADD COLUMN, if the last fields of a clustered index tuple match the 'default row', there is no need to store them. While trimming the entry, we must hold a page latch, so that the table cannot be emptied and the 'default row' be deleted. btr_cur_optimistic_update(), btr_cur_pessimistic_update(), row_upd_clust_rec_by_insert(), row_ins_clust_index_entry_low(): Invoke dtuple_t::trim() if needed. row_ins_clust_index_entry(): Restore dtuple_t::n_fields after calling row_ins_clust_index_entry_low(). rec_get_converted_size(), rec_get_converted_size_comp(): Allow the number of fields to be between n_core_fields and n_fields. Do not support infimum,supremum. They are never supposed to be stored in dtuple_t, because page creation nowadays uses a lower-level method for initializing them. rec_convert_dtuple_to_rec_comp(): Assign the status bits based on the number of fields. btr_cur_trim(): In an update, trim the index entry as needed. For the 'default row', handle rollback specially. For user records, omit fields that match the 'default row'. btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete(): Skip locking and adaptive hash index for the 'default row'. row_log_table_apply_convert_mrec(): Replace 'default row' values if needed. In the temporary file that is applied by row_log_table_apply(), we must identify whether the records contain the extra header for instantly added columns. For now, we will allocate an additional byte for this for ROW_T_INSERT and ROW_T_UPDATE records when the source table has been subject to instant ADD COLUMN. The ROW_T_DELETE records are fine, as they will be converted and will only contain 'core' columns (PRIMARY KEY and some system columns) that are converted from dtuple_t. rec_get_converted_size_temp(), rec_init_offsets_temp(), rec_convert_dtuple_to_temp(): Add the parameter 'status'. REC_INFO_DEFAULT_ROW = REC_INFO_MIN_REC_FLAG | REC_STATUS_COLUMNS_ADDED: An info_bits constant for distinguishing the 'default row' record. rec_comp_status_t: An enum of the status bit values. rec_leaf_format: An enum that replaces the bool parameter of rec_init_offsets_comp_ordinary().
8 years ago
MDEV-15662 Instant DROP COLUMN or changing the order of columns Allow ADD COLUMN anywhere in a table, not only adding as the last column. Allow instant DROP COLUMN and instant changing the order of columns. The added columns will always be added last in clustered index records. In new records, instantly dropped columns will be stored as NULL or empty when possible. Information about dropped and reordered columns will be written in a metadata BLOB (mblob), which is stored before the first 'user' field in the hidden metadata record at the start of the clustered index. The presence of mblob is indicated by setting the delete-mark flag in the metadata record. The metadata BLOB stores the number of clustered index fields, followed by an array of column information for each field. For dropped columns, we store the NOT NULL flag, the fixed length, and for variable-length columns, whether the maximum length exceeded 255 bytes. For non-dropped columns, we store the column position. Unlike with MDEV-11369, when a table becomes empty, it cannot be converted back to the canonical format. The reason for this is that other threads may hold cached objects such as row_prebuilt_t::ins_node that could refer to dropped or reordered index fields. For instant DROP COLUMN and ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC, we must store the n_core_null_bytes in the root page, so that the chain of node pointer records can be followed in order to reach the leftmost leaf page where the metadata record is located. If the mblob is present, we will zero-initialize the strings "infimum" and "supremum" in the root page, and use the last byte of "supremum" for storing the number of null bytes (which are allocated but useless on node pointer pages). This is necessary for btr_cur_instant_init_metadata() to be able to navigate to the mblob. If the PRIMARY KEY contains any variable-length column and some nullable columns were instantly dropped, the dict_index_t::n_nullable in the data dictionary could be smaller than it actually is in the non-leaf pages. Because of this, the non-leaf pages could use more bytes for the null flags than the data dictionary expects, and we could be reading the lengths of the variable-length columns from the wrong offset, and thus reading the child page number from wrong place. This is the result of two design mistakes that involve unnecessary storage of data: First, it is nonsense to store any data fields for the leftmost node pointer records, because the comparisons would be resolved by the MIN_REC_FLAG alone. Second, there cannot be any null fields in the clustered index node pointer fields, but we nevertheless reserve space for all the null flags. Limitations (future work): MDEV-17459 Allow instant ALTER TABLE even if FULLTEXT INDEX exists MDEV-17468 Avoid table rebuild on operations on generated columns MDEV-17494 Refuse ALGORITHM=INSTANT when the row size is too large btr_page_reorganize_low(): Preserve any metadata in the root page. Call lock_move_reorganize_page() only after restoring the "infimum" and "supremum" records, to avoid a memcmp() assertion failure. dict_col_t::DROPPED: Magic value for dict_col_t::ind. dict_col_t::clear_instant(): Renamed from dict_col_t::remove_instant(). Do not assert that the column was instantly added, because we sometimes call this unconditionally for all columns. Convert an instantly added column to a "core column". The old name remove_instant() could be mistaken to refer to "instant DROP COLUMN". dict_col_t::is_added(): Rename from dict_col_t::is_instant(). dtype_t::metadata_blob_init(): Initialize the mblob data type. dtuple_t::is_metadata(), dtuple_t::is_alter_metadata(), upd_t::is_metadata(), upd_t::is_alter_metadata(): Check if info_bits refer to a metadata record. dict_table_t::instant: Metadata about dropped or reordered columns. dict_table_t::prepare_instant(): Prepare ha_innobase_inplace_ctx::instant_table for instant ALTER TABLE. innobase_instant_try() will pass this to dict_table_t::instant_column(). On rollback, dict_table_t::rollback_instant() will be called. dict_table_t::instant_column(): Renamed from instant_add_column(). Add the parameter col_map so that columns can be reordered. Copy and adjust v_cols[] as well. dict_table_t::find(): Find an old column based on a new column number. dict_table_t::serialise_columns(), dict_table_t::deserialise_columns(): Convert the mblob. dict_index_t::instant_metadata(): Create the metadata record for instant ALTER TABLE. Invoke dict_table_t::serialise_columns(). dict_index_t::reconstruct_fields(): Invoked by dict_table_t::deserialise_columns(). dict_index_t::clear_instant_alter(): Move the fields for the dropped columns to the end, and sort the surviving index fields in ascending order of column position. ha_innobase::check_if_supported_inplace_alter(): Do not allow adding a FTS_DOC_ID column if a hidden FTS_DOC_ID column exists due to FULLTEXT INDEX. (This always required ALGORITHM=COPY.) instant_alter_column_possible(): Add a parameter for InnoDB table, to check for additional conditions, such as the maximum number of index fields. ha_innobase_inplace_ctx::first_alter_pos: The first column whose position is affected by instant ADD, DROP, or changing the order of columns. innobase_build_col_map(): Skip added virtual columns. prepare_inplace_add_virtual(): Correctly compute num_to_add_vcol. Remove some unnecessary code. Note that the call to innodb_base_col_setup() should be executed later. commit_try_norebuild(): If ctx->is_instant(), let the virtual columns be added or dropped by innobase_instant_try(). innobase_instant_try(): Fill in a zero default value for the hidden column FTS_DOC_ID (to reduce the work needed in MDEV-17459). If any columns were dropped or reordered (or added not last), delete any SYS_COLUMNS records for the following columns, and insert SYS_COLUMNS records for all subsequent stored columns as well as for all virtual columns. If any virtual column is dropped, rewrite all virtual column metadata. Use a shortcut only for adding virtual columns. This is because innobase_drop_virtual_try() assumes that the dropped virtual columns still exist in ctx->old_table. innodb_update_cols(): Renamed from innodb_update_n_cols(). innobase_add_one_virtual(), innobase_insert_sys_virtual(): Change the return type to bool, and invoke my_error() when detecting an error. innodb_insert_sys_columns(): Insert a record into SYS_COLUMNS. Refactored from innobase_add_one_virtual() and innobase_instant_add_col(). innobase_instant_add_col(): Replace the parameter dfield with type. innobase_instant_drop_cols(): Drop matching columns from SYS_COLUMNS and all columns from SYS_VIRTUAL. innobase_add_virtual_try(), innobase_drop_virtual_try(): Let the caller invoke innodb_update_cols(). innobase_rename_column_try(): Skip dropped columns. commit_cache_norebuild(): Update table->fts->doc_col. dict_mem_table_col_rename_low(): Skip dropped columns. trx_undo_rec_get_partial_row(): Skip dropped columns. trx_undo_update_rec_get_update(): Handle the metadata BLOB correctly. trx_undo_page_report_modify(): Avoid out-of-bounds access to record fields. Log metadata records consistently. Apparently, the first fields of a clustered index may be updated in an update_undo vector when the index is ID_IND of SYS_FOREIGN, as part of renaming the table during ALTER TABLE. Normally, updates of the PRIMARY KEY should be logged as delete-mark and an insert. row_undo_mod_parse_undo_rec(), row_purge_parse_undo_rec(): Use trx_undo_metadata. row_undo_mod_clust_low(): On metadata rollback, roll back the root page too. row_undo_mod_clust(): Relax an assertion. The delete-mark flag was repurposed for ALTER TABLE metadata records. row_rec_to_index_entry_impl(): Add the template parameter mblob and the optional parameter info_bits for specifying the desired new info bits. For the metadata tuple, allow conversion between the original format (ADD COLUMN only) and the generic format (with hidden BLOB). Add the optional parameter "pad" to determine whether the tuple should be padded to the index fields (on ALTER TABLE it should), or whether it should remain at its original size (on rollback). row_build_index_entry_low(): Clean up the code, removing redundant variables and conditions. For instantly dropped columns, generate a dummy value that is NULL, the empty string, or a fixed length of NUL bytes, depending on the type of the dropped column. row_upd_clust_rec_by_insert_inherit_func(): On the update of PRIMARY KEY of a record that contained a dropped column whose value was stored externally, we will be inserting a dummy NULL or empty string value to the field of the dropped column. The externally stored column would eventually be dropped when purge removes the delete-marked record for the old PRIMARY KEY value. btr_index_rec_validate(): Recognize the metadata record. btr_discard_only_page_on_level(): Preserve the generic instant ALTER TABLE metadata. btr_set_instant(): Replaces page_set_instant(). This sets a clustered index root page to the appropriate format, or upgrades from the MDEV-11369 instant ADD COLUMN to generic ALTER TABLE format. btr_cur_instant_init_low(): Read and validate the metadata BLOB page before reconstructing the dictionary information based on it. btr_cur_instant_init_metadata(): Do not read any lengths from the metadata record header before reading the BLOB. At this point, we would not actually know how many nullable fields the metadata record contains. btr_cur_instant_root_init(): Initialize n_core_null_bytes in one of two possible ways. btr_cur_trim(): Handle the mblob record. row_metadata_to_tuple(): Convert a metadata record to a data tuple, based on the new info_bits of the metadata record. btr_cur_pessimistic_update(): Invoke row_metadata_to_tuple() if needed. Invoke dtuple_convert_big_rec() for metadata records if the record is too large, or if the mblob is not yet marked as externally stored. btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete(): When the last user record is deleted, do not delete the generic instant ALTER TABLE metadata record. Only delete MDEV-11369 instant ADD COLUMN metadata records. btr_cur_optimistic_insert(): Avoid unnecessary computation of rec_size. btr_pcur_store_position(): Allow a logically empty page to contain a metadata record for generic ALTER TABLE. REC_INFO_DEFAULT_ROW_ADD: Renamed from REC_INFO_DEFAULT_ROW. This is for the old instant ADD COLUMN (MDEV-11369) only. REC_INFO_DEFAULT_ROW_ALTER: The more generic metadata record, with additional information for dropped or reordered columns. rec_info_bits_valid(): Remove. The only case when this would fail is when the record is the generic ALTER TABLE metadata record. rec_is_alter_metadata(): Check if a record is the metadata record for instant ALTER TABLE (other than ADD COLUMN). NOTE: This function must not be invoked on node pointer records, because the delete-mark flag in those records may be set (it is garbage), and then a debug assertion could fail because index->is_instant() does not necessarily hold. rec_is_add_metadata(): Check if a record is MDEV-11369 ADD COLUMN metadata record (not more generic instant ALTER TABLE). rec_get_converted_size_comp_prefix_low(): Assume that the metadata field will be stored externally. In dtuple_convert_big_rec() during the rec_get_converted_size() call, it would not be there yet. rec_get_converted_size_comp(): Replace status,fields,n_fields with tuple. rec_init_offsets_comp_ordinary(), rec_get_converted_size_comp_prefix_low(), rec_convert_dtuple_to_rec_comp(): Add template<bool mblob = false>. With mblob=true, process a record with a metadata BLOB. rec_copy_prefix_to_buf(): Assert that no fields beyond the key and system columns are being copied. Exclude the metadata BLOB field. rec_convert_dtuple_to_metadata_comp(): Convert an alter metadata tuple into a record. row_upd_index_replace_metadata(): Apply an update vector to an alter_metadata tuple. row_log_allocate(): Replace dict_index_t::is_instant() with a more appropriate condition that ignores dict_table_t::instant. Only a table on which the MDEV-11369 ADD COLUMN was performed can "lose its instantness" when it becomes empty. After instant DROP COLUMN or reordering columns, we cannot simply convert the table to the canonical format, because the data dictionary cache and all possibly existing references to it from other client connection threads would have to be adjusted. row_quiesce_write_index_fields(): Do not crash when the table contains an instantly dropped column. Thanks to Thirunarayanan Balathandayuthapani for discussing the design and implementing an initial prototype of this. Thanks to Matthias Leich for testing.
7 years ago
MDEV-11369 Instant ADD COLUMN for InnoDB For InnoDB tables, adding, dropping and reordering columns has required a rebuild of the table and all its indexes. Since MySQL 5.6 (and MariaDB 10.0) this has been supported online (LOCK=NONE), allowing concurrent modification of the tables. This work revises the InnoDB ROW_FORMAT=REDUNDANT, ROW_FORMAT=COMPACT and ROW_FORMAT=DYNAMIC so that columns can be appended instantaneously, with only minor changes performed to the table structure. The counter innodb_instant_alter_column in INFORMATION_SCHEMA.GLOBAL_STATUS is incremented whenever a table rebuild operation is converted into an instant ADD COLUMN operation. ROW_FORMAT=COMPRESSED tables will not support instant ADD COLUMN. Some usability limitations will be addressed in subsequent work: MDEV-13134 Introduce ALTER TABLE attributes ALGORITHM=NOCOPY and ALGORITHM=INSTANT MDEV-14016 Allow instant ADD COLUMN, ADD INDEX, LOCK=NONE The format of the clustered index (PRIMARY KEY) is changed as follows: (1) The FIL_PAGE_TYPE of the root page will be FIL_PAGE_TYPE_INSTANT, and a new field PAGE_INSTANT will contain the original number of fields in the clustered index ('core' fields). If instant ADD COLUMN has not been used or the table becomes empty, or the very first instant ADD COLUMN operation is rolled back, the fields PAGE_INSTANT and FIL_PAGE_TYPE will be reset to 0 and FIL_PAGE_INDEX. (2) A special 'default row' record is inserted into the leftmost leaf, between the page infimum and the first user record. This record is distinguished by the REC_INFO_MIN_REC_FLAG, and it is otherwise in the same format as records that contain values for the instantly added columns. This 'default row' always has the same number of fields as the clustered index according to the table definition. The values of 'core' fields are to be ignored. For other fields, the 'default row' will contain the default values as they were during the ALTER TABLE statement. (If the column default values are changed later, those values will only be stored in the .frm file. The 'default row' will contain the original evaluated values, which must be the same for every row.) The 'default row' must be completely hidden from higher-level access routines. Assertions have been added to ensure that no 'default row' is ever present in the adaptive hash index or in locked records. The 'default row' is never delete-marked. (3) In clustered index leaf page records, the number of fields must reside between the number of 'core' fields (dict_index_t::n_core_fields introduced in this work) and dict_index_t::n_fields. If the number of fields is less than dict_index_t::n_fields, the missing fields are replaced with the column value of the 'default row'. Note: The number of fields in the record may shrink if some of the last instantly added columns are updated to the value that is in the 'default row'. The function btr_cur_trim() implements this 'compression' on update and rollback; dtuple::trim() implements it on insert. (4) In ROW_FORMAT=COMPACT and ROW_FORMAT=DYNAMIC records, the new status value REC_STATUS_COLUMNS_ADDED will indicate the presence of a new record header that will encode n_fields-n_core_fields-1 in 1 or 2 bytes. (In ROW_FORMAT=REDUNDANT records, the record header always explicitly encodes the number of fields.) We introduce the undo log record type TRX_UNDO_INSERT_DEFAULT for covering the insert of the 'default row' record when instant ADD COLUMN is used for the first time. Subsequent instant ADD COLUMN can use TRX_UNDO_UPD_EXIST_REC. This is joint work with Vin Chen (陈福荣) from Tencent. The design that was discussed in April 2017 would not have allowed import or export of data files, because instead of the 'default row' it would have introduced a data dictionary table. The test rpl.rpl_alter_instant is exactly as contributed in pull request #408. The test innodb.instant_alter is based on a contributed test. The redo log record format changes for ROW_FORMAT=DYNAMIC and ROW_FORMAT=COMPACT are as contributed. (With this change present, crash recovery from MariaDB 10.3.1 will fail in spectacular ways!) Also the semantics of higher-level redo log records that modify the PAGE_INSTANT field is changed. The redo log format version identifier was already changed to LOG_HEADER_FORMAT_CURRENT=103 in MariaDB 10.3.1. Everything else has been rewritten by me. Thanks to Elena Stepanova, the code has been tested extensively. When rolling back an instant ADD COLUMN operation, we must empty the PAGE_FREE list after deleting or shortening the 'default row' record, by calling either btr_page_empty() or btr_page_reorganize(). We must know the size of each entry in the PAGE_FREE list. If rollback left a freed copy of the 'default row' in the PAGE_FREE list, we would be unable to determine its size (if it is in ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC) because it would contain more fields than the rolled-back definition of the clustered index. UNIV_SQL_DEFAULT: A new special constant that designates an instantly added column that is not present in the clustered index record. len_is_stored(): Check if a length is an actual length. There are two magic length values: UNIV_SQL_DEFAULT, UNIV_SQL_NULL. dict_col_t::def_val: The 'default row' value of the column. If the column is not added instantly, def_val.len will be UNIV_SQL_DEFAULT. dict_col_t: Add the accessors is_virtual(), is_nullable(), is_instant(), instant_value(). dict_col_t::remove_instant(): Remove the 'instant ADD' status of a column. dict_col_t::name(const dict_table_t& table): Replaces dict_table_get_col_name(). dict_index_t::n_core_fields: The original number of fields. For secondary indexes and if instant ADD COLUMN has not been used, this will be equal to dict_index_t::n_fields. dict_index_t::n_core_null_bytes: Number of bytes needed to represent the null flags; usually equal to UT_BITS_IN_BYTES(n_nullable). dict_index_t::NO_CORE_NULL_BYTES: Magic value signalling that n_core_null_bytes was not initialized yet from the clustered index root page. dict_index_t: Add the accessors is_instant(), is_clust(), get_n_nullable(), instant_field_value(). dict_index_t::instant_add_field(): Adjust clustered index metadata for instant ADD COLUMN. dict_index_t::remove_instant(): Remove the 'instant ADD' status of a clustered index when the table becomes empty, or the very first instant ADD COLUMN operation is rolled back. dict_table_t: Add the accessors is_instant(), is_temporary(), supports_instant(). dict_table_t::instant_add_column(): Adjust metadata for instant ADD COLUMN. dict_table_t::rollback_instant(): Adjust metadata on the rollback of instant ADD COLUMN. prepare_inplace_alter_table_dict(): First create the ctx->new_table, and only then decide if the table really needs to be rebuilt. We must split the creation of table or index metadata from the creation of the dictionary table records and the creation of the data. In this way, we can transform a table-rebuilding operation into an instant ADD COLUMN operation. Dictionary objects will only be added to cache when table rebuilding or index creation is needed. The ctx->instant_table will never be added to cache. dict_table_t::add_to_cache(): Modified and renamed from dict_table_add_to_cache(). Do not modify the table metadata. Let the callers invoke dict_table_add_system_columns() and if needed, set can_be_evicted. dict_create_sys_tables_tuple(), dict_create_table_step(): Omit the system columns (which will now exist in the dict_table_t object already at this point). dict_create_table_step(): Expect the callers to invoke dict_table_add_system_columns(). pars_create_table(): Before creating the table creation execution graph, invoke dict_table_add_system_columns(). row_create_table_for_mysql(): Expect all callers to invoke dict_table_add_system_columns(). create_index_dict(): Replaces row_merge_create_index_graph(). innodb_update_n_cols(): Renamed from innobase_update_n_virtual(). Call my_error() if an error occurs. btr_cur_instant_init(), btr_cur_instant_init_low(), btr_cur_instant_root_init(): Load additional metadata from the clustered index and set dict_index_t::n_core_null_bytes. This is invoked when table metadata is first loaded into the data dictionary. dict_boot(): Initialize n_core_null_bytes for the four hard-coded dictionary tables. dict_create_index_step(): Initialize n_core_null_bytes. This is executed as part of CREATE TABLE. dict_index_build_internal_clust(): Initialize n_core_null_bytes to NO_CORE_NULL_BYTES if table->supports_instant(). row_create_index_for_mysql(): Initialize n_core_null_bytes for CREATE TEMPORARY TABLE. commit_cache_norebuild(): Call the code to rename or enlarge columns in the cache only if instant ADD COLUMN is not being used. (Instant ADD COLUMN would copy all column metadata from instant_table to old_table, including the names and lengths.) PAGE_INSTANT: A new 13-bit field for storing dict_index_t::n_core_fields. This is repurposing the 16-bit field PAGE_DIRECTION, of which only the least significant 3 bits were used. The original byte containing PAGE_DIRECTION will be accessible via the new constant PAGE_DIRECTION_B. page_get_instant(), page_set_instant(): Accessors for the PAGE_INSTANT. page_ptr_get_direction(), page_get_direction(), page_ptr_set_direction(): Accessors for PAGE_DIRECTION. page_direction_reset(): Reset PAGE_DIRECTION, PAGE_N_DIRECTION. page_direction_increment(): Increment PAGE_N_DIRECTION and set PAGE_DIRECTION. rec_get_offsets(): Use the 'leaf' parameter for non-debug purposes, and assume that heap_no is always set. Initialize all dict_index_t::n_fields for ROW_FORMAT=REDUNDANT records, even if the record contains fewer fields. rec_offs_make_valid(): Add the parameter 'leaf'. rec_copy_prefix_to_dtuple(): Assert that the tuple is only built on the core fields. Instant ADD COLUMN only applies to the clustered index, and we should never build a search key that has more than the PRIMARY KEY and possibly DB_TRX_ID,DB_ROLL_PTR. All these columns are always present. dict_index_build_data_tuple(): Remove assertions that would be duplicated in rec_copy_prefix_to_dtuple(). rec_init_offsets(): Support ROW_FORMAT=REDUNDANT records whose number of fields is between n_core_fields and n_fields. cmp_rec_rec_with_match(): Implement the comparison between two MIN_REC_FLAG records. trx_t::in_rollback: Make the field available in non-debug builds. trx_start_for_ddl_low(): Remove dangerous error-tolerance. A dictionary transaction must be flagged as such before it has generated any undo log records. This is because trx_undo_assign_undo() will mark the transaction as a dictionary transaction in the undo log header right before the very first undo log record is being written. btr_index_rec_validate(): Account for instant ADD COLUMN row_undo_ins_remove_clust_rec(): On the rollback of an insert into SYS_COLUMNS, revert instant ADD COLUMN in the cache by removing the last column from the table and the clustered index. row_search_on_row_ref(), row_undo_mod_parse_undo_rec(), row_undo_mod(), trx_undo_update_rec_get_update(): Handle the 'default row' as a special case. dtuple_t::trim(index): Omit a redundant suffix of an index tuple right before insert or update. After instant ADD COLUMN, if the last fields of a clustered index tuple match the 'default row', there is no need to store them. While trimming the entry, we must hold a page latch, so that the table cannot be emptied and the 'default row' be deleted. btr_cur_optimistic_update(), btr_cur_pessimistic_update(), row_upd_clust_rec_by_insert(), row_ins_clust_index_entry_low(): Invoke dtuple_t::trim() if needed. row_ins_clust_index_entry(): Restore dtuple_t::n_fields after calling row_ins_clust_index_entry_low(). rec_get_converted_size(), rec_get_converted_size_comp(): Allow the number of fields to be between n_core_fields and n_fields. Do not support infimum,supremum. They are never supposed to be stored in dtuple_t, because page creation nowadays uses a lower-level method for initializing them. rec_convert_dtuple_to_rec_comp(): Assign the status bits based on the number of fields. btr_cur_trim(): In an update, trim the index entry as needed. For the 'default row', handle rollback specially. For user records, omit fields that match the 'default row'. btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete(): Skip locking and adaptive hash index for the 'default row'. row_log_table_apply_convert_mrec(): Replace 'default row' values if needed. In the temporary file that is applied by row_log_table_apply(), we must identify whether the records contain the extra header for instantly added columns. For now, we will allocate an additional byte for this for ROW_T_INSERT and ROW_T_UPDATE records when the source table has been subject to instant ADD COLUMN. The ROW_T_DELETE records are fine, as they will be converted and will only contain 'core' columns (PRIMARY KEY and some system columns) that are converted from dtuple_t. rec_get_converted_size_temp(), rec_init_offsets_temp(), rec_convert_dtuple_to_temp(): Add the parameter 'status'. REC_INFO_DEFAULT_ROW = REC_INFO_MIN_REC_FLAG | REC_STATUS_COLUMNS_ADDED: An info_bits constant for distinguishing the 'default row' record. rec_comp_status_t: An enum of the status bit values. rec_leaf_format: An enum that replaces the bool parameter of rec_init_offsets_comp_ordinary().
8 years ago
MDEV-15662 Instant DROP COLUMN or changing the order of columns Allow ADD COLUMN anywhere in a table, not only adding as the last column. Allow instant DROP COLUMN and instant changing the order of columns. The added columns will always be added last in clustered index records. In new records, instantly dropped columns will be stored as NULL or empty when possible. Information about dropped and reordered columns will be written in a metadata BLOB (mblob), which is stored before the first 'user' field in the hidden metadata record at the start of the clustered index. The presence of mblob is indicated by setting the delete-mark flag in the metadata record. The metadata BLOB stores the number of clustered index fields, followed by an array of column information for each field. For dropped columns, we store the NOT NULL flag, the fixed length, and for variable-length columns, whether the maximum length exceeded 255 bytes. For non-dropped columns, we store the column position. Unlike with MDEV-11369, when a table becomes empty, it cannot be converted back to the canonical format. The reason for this is that other threads may hold cached objects such as row_prebuilt_t::ins_node that could refer to dropped or reordered index fields. For instant DROP COLUMN and ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC, we must store the n_core_null_bytes in the root page, so that the chain of node pointer records can be followed in order to reach the leftmost leaf page where the metadata record is located. If the mblob is present, we will zero-initialize the strings "infimum" and "supremum" in the root page, and use the last byte of "supremum" for storing the number of null bytes (which are allocated but useless on node pointer pages). This is necessary for btr_cur_instant_init_metadata() to be able to navigate to the mblob. If the PRIMARY KEY contains any variable-length column and some nullable columns were instantly dropped, the dict_index_t::n_nullable in the data dictionary could be smaller than it actually is in the non-leaf pages. Because of this, the non-leaf pages could use more bytes for the null flags than the data dictionary expects, and we could be reading the lengths of the variable-length columns from the wrong offset, and thus reading the child page number from wrong place. This is the result of two design mistakes that involve unnecessary storage of data: First, it is nonsense to store any data fields for the leftmost node pointer records, because the comparisons would be resolved by the MIN_REC_FLAG alone. Second, there cannot be any null fields in the clustered index node pointer fields, but we nevertheless reserve space for all the null flags. Limitations (future work): MDEV-17459 Allow instant ALTER TABLE even if FULLTEXT INDEX exists MDEV-17468 Avoid table rebuild on operations on generated columns MDEV-17494 Refuse ALGORITHM=INSTANT when the row size is too large btr_page_reorganize_low(): Preserve any metadata in the root page. Call lock_move_reorganize_page() only after restoring the "infimum" and "supremum" records, to avoid a memcmp() assertion failure. dict_col_t::DROPPED: Magic value for dict_col_t::ind. dict_col_t::clear_instant(): Renamed from dict_col_t::remove_instant(). Do not assert that the column was instantly added, because we sometimes call this unconditionally for all columns. Convert an instantly added column to a "core column". The old name remove_instant() could be mistaken to refer to "instant DROP COLUMN". dict_col_t::is_added(): Rename from dict_col_t::is_instant(). dtype_t::metadata_blob_init(): Initialize the mblob data type. dtuple_t::is_metadata(), dtuple_t::is_alter_metadata(), upd_t::is_metadata(), upd_t::is_alter_metadata(): Check if info_bits refer to a metadata record. dict_table_t::instant: Metadata about dropped or reordered columns. dict_table_t::prepare_instant(): Prepare ha_innobase_inplace_ctx::instant_table for instant ALTER TABLE. innobase_instant_try() will pass this to dict_table_t::instant_column(). On rollback, dict_table_t::rollback_instant() will be called. dict_table_t::instant_column(): Renamed from instant_add_column(). Add the parameter col_map so that columns can be reordered. Copy and adjust v_cols[] as well. dict_table_t::find(): Find an old column based on a new column number. dict_table_t::serialise_columns(), dict_table_t::deserialise_columns(): Convert the mblob. dict_index_t::instant_metadata(): Create the metadata record for instant ALTER TABLE. Invoke dict_table_t::serialise_columns(). dict_index_t::reconstruct_fields(): Invoked by dict_table_t::deserialise_columns(). dict_index_t::clear_instant_alter(): Move the fields for the dropped columns to the end, and sort the surviving index fields in ascending order of column position. ha_innobase::check_if_supported_inplace_alter(): Do not allow adding a FTS_DOC_ID column if a hidden FTS_DOC_ID column exists due to FULLTEXT INDEX. (This always required ALGORITHM=COPY.) instant_alter_column_possible(): Add a parameter for InnoDB table, to check for additional conditions, such as the maximum number of index fields. ha_innobase_inplace_ctx::first_alter_pos: The first column whose position is affected by instant ADD, DROP, or changing the order of columns. innobase_build_col_map(): Skip added virtual columns. prepare_inplace_add_virtual(): Correctly compute num_to_add_vcol. Remove some unnecessary code. Note that the call to innodb_base_col_setup() should be executed later. commit_try_norebuild(): If ctx->is_instant(), let the virtual columns be added or dropped by innobase_instant_try(). innobase_instant_try(): Fill in a zero default value for the hidden column FTS_DOC_ID (to reduce the work needed in MDEV-17459). If any columns were dropped or reordered (or added not last), delete any SYS_COLUMNS records for the following columns, and insert SYS_COLUMNS records for all subsequent stored columns as well as for all virtual columns. If any virtual column is dropped, rewrite all virtual column metadata. Use a shortcut only for adding virtual columns. This is because innobase_drop_virtual_try() assumes that the dropped virtual columns still exist in ctx->old_table. innodb_update_cols(): Renamed from innodb_update_n_cols(). innobase_add_one_virtual(), innobase_insert_sys_virtual(): Change the return type to bool, and invoke my_error() when detecting an error. innodb_insert_sys_columns(): Insert a record into SYS_COLUMNS. Refactored from innobase_add_one_virtual() and innobase_instant_add_col(). innobase_instant_add_col(): Replace the parameter dfield with type. innobase_instant_drop_cols(): Drop matching columns from SYS_COLUMNS and all columns from SYS_VIRTUAL. innobase_add_virtual_try(), innobase_drop_virtual_try(): Let the caller invoke innodb_update_cols(). innobase_rename_column_try(): Skip dropped columns. commit_cache_norebuild(): Update table->fts->doc_col. dict_mem_table_col_rename_low(): Skip dropped columns. trx_undo_rec_get_partial_row(): Skip dropped columns. trx_undo_update_rec_get_update(): Handle the metadata BLOB correctly. trx_undo_page_report_modify(): Avoid out-of-bounds access to record fields. Log metadata records consistently. Apparently, the first fields of a clustered index may be updated in an update_undo vector when the index is ID_IND of SYS_FOREIGN, as part of renaming the table during ALTER TABLE. Normally, updates of the PRIMARY KEY should be logged as delete-mark and an insert. row_undo_mod_parse_undo_rec(), row_purge_parse_undo_rec(): Use trx_undo_metadata. row_undo_mod_clust_low(): On metadata rollback, roll back the root page too. row_undo_mod_clust(): Relax an assertion. The delete-mark flag was repurposed for ALTER TABLE metadata records. row_rec_to_index_entry_impl(): Add the template parameter mblob and the optional parameter info_bits for specifying the desired new info bits. For the metadata tuple, allow conversion between the original format (ADD COLUMN only) and the generic format (with hidden BLOB). Add the optional parameter "pad" to determine whether the tuple should be padded to the index fields (on ALTER TABLE it should), or whether it should remain at its original size (on rollback). row_build_index_entry_low(): Clean up the code, removing redundant variables and conditions. For instantly dropped columns, generate a dummy value that is NULL, the empty string, or a fixed length of NUL bytes, depending on the type of the dropped column. row_upd_clust_rec_by_insert_inherit_func(): On the update of PRIMARY KEY of a record that contained a dropped column whose value was stored externally, we will be inserting a dummy NULL or empty string value to the field of the dropped column. The externally stored column would eventually be dropped when purge removes the delete-marked record for the old PRIMARY KEY value. btr_index_rec_validate(): Recognize the metadata record. btr_discard_only_page_on_level(): Preserve the generic instant ALTER TABLE metadata. btr_set_instant(): Replaces page_set_instant(). This sets a clustered index root page to the appropriate format, or upgrades from the MDEV-11369 instant ADD COLUMN to generic ALTER TABLE format. btr_cur_instant_init_low(): Read and validate the metadata BLOB page before reconstructing the dictionary information based on it. btr_cur_instant_init_metadata(): Do not read any lengths from the metadata record header before reading the BLOB. At this point, we would not actually know how many nullable fields the metadata record contains. btr_cur_instant_root_init(): Initialize n_core_null_bytes in one of two possible ways. btr_cur_trim(): Handle the mblob record. row_metadata_to_tuple(): Convert a metadata record to a data tuple, based on the new info_bits of the metadata record. btr_cur_pessimistic_update(): Invoke row_metadata_to_tuple() if needed. Invoke dtuple_convert_big_rec() for metadata records if the record is too large, or if the mblob is not yet marked as externally stored. btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete(): When the last user record is deleted, do not delete the generic instant ALTER TABLE metadata record. Only delete MDEV-11369 instant ADD COLUMN metadata records. btr_cur_optimistic_insert(): Avoid unnecessary computation of rec_size. btr_pcur_store_position(): Allow a logically empty page to contain a metadata record for generic ALTER TABLE. REC_INFO_DEFAULT_ROW_ADD: Renamed from REC_INFO_DEFAULT_ROW. This is for the old instant ADD COLUMN (MDEV-11369) only. REC_INFO_DEFAULT_ROW_ALTER: The more generic metadata record, with additional information for dropped or reordered columns. rec_info_bits_valid(): Remove. The only case when this would fail is when the record is the generic ALTER TABLE metadata record. rec_is_alter_metadata(): Check if a record is the metadata record for instant ALTER TABLE (other than ADD COLUMN). NOTE: This function must not be invoked on node pointer records, because the delete-mark flag in those records may be set (it is garbage), and then a debug assertion could fail because index->is_instant() does not necessarily hold. rec_is_add_metadata(): Check if a record is MDEV-11369 ADD COLUMN metadata record (not more generic instant ALTER TABLE). rec_get_converted_size_comp_prefix_low(): Assume that the metadata field will be stored externally. In dtuple_convert_big_rec() during the rec_get_converted_size() call, it would not be there yet. rec_get_converted_size_comp(): Replace status,fields,n_fields with tuple. rec_init_offsets_comp_ordinary(), rec_get_converted_size_comp_prefix_low(), rec_convert_dtuple_to_rec_comp(): Add template<bool mblob = false>. With mblob=true, process a record with a metadata BLOB. rec_copy_prefix_to_buf(): Assert that no fields beyond the key and system columns are being copied. Exclude the metadata BLOB field. rec_convert_dtuple_to_metadata_comp(): Convert an alter metadata tuple into a record. row_upd_index_replace_metadata(): Apply an update vector to an alter_metadata tuple. row_log_allocate(): Replace dict_index_t::is_instant() with a more appropriate condition that ignores dict_table_t::instant. Only a table on which the MDEV-11369 ADD COLUMN was performed can "lose its instantness" when it becomes empty. After instant DROP COLUMN or reordering columns, we cannot simply convert the table to the canonical format, because the data dictionary cache and all possibly existing references to it from other client connection threads would have to be adjusted. row_quiesce_write_index_fields(): Do not crash when the table contains an instantly dropped column. Thanks to Thirunarayanan Balathandayuthapani for discussing the design and implementing an initial prototype of this. Thanks to Matthias Leich for testing.
7 years ago
MDEV-15662 Instant DROP COLUMN or changing the order of columns Allow ADD COLUMN anywhere in a table, not only adding as the last column. Allow instant DROP COLUMN and instant changing the order of columns. The added columns will always be added last in clustered index records. In new records, instantly dropped columns will be stored as NULL or empty when possible. Information about dropped and reordered columns will be written in a metadata BLOB (mblob), which is stored before the first 'user' field in the hidden metadata record at the start of the clustered index. The presence of mblob is indicated by setting the delete-mark flag in the metadata record. The metadata BLOB stores the number of clustered index fields, followed by an array of column information for each field. For dropped columns, we store the NOT NULL flag, the fixed length, and for variable-length columns, whether the maximum length exceeded 255 bytes. For non-dropped columns, we store the column position. Unlike with MDEV-11369, when a table becomes empty, it cannot be converted back to the canonical format. The reason for this is that other threads may hold cached objects such as row_prebuilt_t::ins_node that could refer to dropped or reordered index fields. For instant DROP COLUMN and ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC, we must store the n_core_null_bytes in the root page, so that the chain of node pointer records can be followed in order to reach the leftmost leaf page where the metadata record is located. If the mblob is present, we will zero-initialize the strings "infimum" and "supremum" in the root page, and use the last byte of "supremum" for storing the number of null bytes (which are allocated but useless on node pointer pages). This is necessary for btr_cur_instant_init_metadata() to be able to navigate to the mblob. If the PRIMARY KEY contains any variable-length column and some nullable columns were instantly dropped, the dict_index_t::n_nullable in the data dictionary could be smaller than it actually is in the non-leaf pages. Because of this, the non-leaf pages could use more bytes for the null flags than the data dictionary expects, and we could be reading the lengths of the variable-length columns from the wrong offset, and thus reading the child page number from wrong place. This is the result of two design mistakes that involve unnecessary storage of data: First, it is nonsense to store any data fields for the leftmost node pointer records, because the comparisons would be resolved by the MIN_REC_FLAG alone. Second, there cannot be any null fields in the clustered index node pointer fields, but we nevertheless reserve space for all the null flags. Limitations (future work): MDEV-17459 Allow instant ALTER TABLE even if FULLTEXT INDEX exists MDEV-17468 Avoid table rebuild on operations on generated columns MDEV-17494 Refuse ALGORITHM=INSTANT when the row size is too large btr_page_reorganize_low(): Preserve any metadata in the root page. Call lock_move_reorganize_page() only after restoring the "infimum" and "supremum" records, to avoid a memcmp() assertion failure. dict_col_t::DROPPED: Magic value for dict_col_t::ind. dict_col_t::clear_instant(): Renamed from dict_col_t::remove_instant(). Do not assert that the column was instantly added, because we sometimes call this unconditionally for all columns. Convert an instantly added column to a "core column". The old name remove_instant() could be mistaken to refer to "instant DROP COLUMN". dict_col_t::is_added(): Rename from dict_col_t::is_instant(). dtype_t::metadata_blob_init(): Initialize the mblob data type. dtuple_t::is_metadata(), dtuple_t::is_alter_metadata(), upd_t::is_metadata(), upd_t::is_alter_metadata(): Check if info_bits refer to a metadata record. dict_table_t::instant: Metadata about dropped or reordered columns. dict_table_t::prepare_instant(): Prepare ha_innobase_inplace_ctx::instant_table for instant ALTER TABLE. innobase_instant_try() will pass this to dict_table_t::instant_column(). On rollback, dict_table_t::rollback_instant() will be called. dict_table_t::instant_column(): Renamed from instant_add_column(). Add the parameter col_map so that columns can be reordered. Copy and adjust v_cols[] as well. dict_table_t::find(): Find an old column based on a new column number. dict_table_t::serialise_columns(), dict_table_t::deserialise_columns(): Convert the mblob. dict_index_t::instant_metadata(): Create the metadata record for instant ALTER TABLE. Invoke dict_table_t::serialise_columns(). dict_index_t::reconstruct_fields(): Invoked by dict_table_t::deserialise_columns(). dict_index_t::clear_instant_alter(): Move the fields for the dropped columns to the end, and sort the surviving index fields in ascending order of column position. ha_innobase::check_if_supported_inplace_alter(): Do not allow adding a FTS_DOC_ID column if a hidden FTS_DOC_ID column exists due to FULLTEXT INDEX. (This always required ALGORITHM=COPY.) instant_alter_column_possible(): Add a parameter for InnoDB table, to check for additional conditions, such as the maximum number of index fields. ha_innobase_inplace_ctx::first_alter_pos: The first column whose position is affected by instant ADD, DROP, or changing the order of columns. innobase_build_col_map(): Skip added virtual columns. prepare_inplace_add_virtual(): Correctly compute num_to_add_vcol. Remove some unnecessary code. Note that the call to innodb_base_col_setup() should be executed later. commit_try_norebuild(): If ctx->is_instant(), let the virtual columns be added or dropped by innobase_instant_try(). innobase_instant_try(): Fill in a zero default value for the hidden column FTS_DOC_ID (to reduce the work needed in MDEV-17459). If any columns were dropped or reordered (or added not last), delete any SYS_COLUMNS records for the following columns, and insert SYS_COLUMNS records for all subsequent stored columns as well as for all virtual columns. If any virtual column is dropped, rewrite all virtual column metadata. Use a shortcut only for adding virtual columns. This is because innobase_drop_virtual_try() assumes that the dropped virtual columns still exist in ctx->old_table. innodb_update_cols(): Renamed from innodb_update_n_cols(). innobase_add_one_virtual(), innobase_insert_sys_virtual(): Change the return type to bool, and invoke my_error() when detecting an error. innodb_insert_sys_columns(): Insert a record into SYS_COLUMNS. Refactored from innobase_add_one_virtual() and innobase_instant_add_col(). innobase_instant_add_col(): Replace the parameter dfield with type. innobase_instant_drop_cols(): Drop matching columns from SYS_COLUMNS and all columns from SYS_VIRTUAL. innobase_add_virtual_try(), innobase_drop_virtual_try(): Let the caller invoke innodb_update_cols(). innobase_rename_column_try(): Skip dropped columns. commit_cache_norebuild(): Update table->fts->doc_col. dict_mem_table_col_rename_low(): Skip dropped columns. trx_undo_rec_get_partial_row(): Skip dropped columns. trx_undo_update_rec_get_update(): Handle the metadata BLOB correctly. trx_undo_page_report_modify(): Avoid out-of-bounds access to record fields. Log metadata records consistently. Apparently, the first fields of a clustered index may be updated in an update_undo vector when the index is ID_IND of SYS_FOREIGN, as part of renaming the table during ALTER TABLE. Normally, updates of the PRIMARY KEY should be logged as delete-mark and an insert. row_undo_mod_parse_undo_rec(), row_purge_parse_undo_rec(): Use trx_undo_metadata. row_undo_mod_clust_low(): On metadata rollback, roll back the root page too. row_undo_mod_clust(): Relax an assertion. The delete-mark flag was repurposed for ALTER TABLE metadata records. row_rec_to_index_entry_impl(): Add the template parameter mblob and the optional parameter info_bits for specifying the desired new info bits. For the metadata tuple, allow conversion between the original format (ADD COLUMN only) and the generic format (with hidden BLOB). Add the optional parameter "pad" to determine whether the tuple should be padded to the index fields (on ALTER TABLE it should), or whether it should remain at its original size (on rollback). row_build_index_entry_low(): Clean up the code, removing redundant variables and conditions. For instantly dropped columns, generate a dummy value that is NULL, the empty string, or a fixed length of NUL bytes, depending on the type of the dropped column. row_upd_clust_rec_by_insert_inherit_func(): On the update of PRIMARY KEY of a record that contained a dropped column whose value was stored externally, we will be inserting a dummy NULL or empty string value to the field of the dropped column. The externally stored column would eventually be dropped when purge removes the delete-marked record for the old PRIMARY KEY value. btr_index_rec_validate(): Recognize the metadata record. btr_discard_only_page_on_level(): Preserve the generic instant ALTER TABLE metadata. btr_set_instant(): Replaces page_set_instant(). This sets a clustered index root page to the appropriate format, or upgrades from the MDEV-11369 instant ADD COLUMN to generic ALTER TABLE format. btr_cur_instant_init_low(): Read and validate the metadata BLOB page before reconstructing the dictionary information based on it. btr_cur_instant_init_metadata(): Do not read any lengths from the metadata record header before reading the BLOB. At this point, we would not actually know how many nullable fields the metadata record contains. btr_cur_instant_root_init(): Initialize n_core_null_bytes in one of two possible ways. btr_cur_trim(): Handle the mblob record. row_metadata_to_tuple(): Convert a metadata record to a data tuple, based on the new info_bits of the metadata record. btr_cur_pessimistic_update(): Invoke row_metadata_to_tuple() if needed. Invoke dtuple_convert_big_rec() for metadata records if the record is too large, or if the mblob is not yet marked as externally stored. btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete(): When the last user record is deleted, do not delete the generic instant ALTER TABLE metadata record. Only delete MDEV-11369 instant ADD COLUMN metadata records. btr_cur_optimistic_insert(): Avoid unnecessary computation of rec_size. btr_pcur_store_position(): Allow a logically empty page to contain a metadata record for generic ALTER TABLE. REC_INFO_DEFAULT_ROW_ADD: Renamed from REC_INFO_DEFAULT_ROW. This is for the old instant ADD COLUMN (MDEV-11369) only. REC_INFO_DEFAULT_ROW_ALTER: The more generic metadata record, with additional information for dropped or reordered columns. rec_info_bits_valid(): Remove. The only case when this would fail is when the record is the generic ALTER TABLE metadata record. rec_is_alter_metadata(): Check if a record is the metadata record for instant ALTER TABLE (other than ADD COLUMN). NOTE: This function must not be invoked on node pointer records, because the delete-mark flag in those records may be set (it is garbage), and then a debug assertion could fail because index->is_instant() does not necessarily hold. rec_is_add_metadata(): Check if a record is MDEV-11369 ADD COLUMN metadata record (not more generic instant ALTER TABLE). rec_get_converted_size_comp_prefix_low(): Assume that the metadata field will be stored externally. In dtuple_convert_big_rec() during the rec_get_converted_size() call, it would not be there yet. rec_get_converted_size_comp(): Replace status,fields,n_fields with tuple. rec_init_offsets_comp_ordinary(), rec_get_converted_size_comp_prefix_low(), rec_convert_dtuple_to_rec_comp(): Add template<bool mblob = false>. With mblob=true, process a record with a metadata BLOB. rec_copy_prefix_to_buf(): Assert that no fields beyond the key and system columns are being copied. Exclude the metadata BLOB field. rec_convert_dtuple_to_metadata_comp(): Convert an alter metadata tuple into a record. row_upd_index_replace_metadata(): Apply an update vector to an alter_metadata tuple. row_log_allocate(): Replace dict_index_t::is_instant() with a more appropriate condition that ignores dict_table_t::instant. Only a table on which the MDEV-11369 ADD COLUMN was performed can "lose its instantness" when it becomes empty. After instant DROP COLUMN or reordering columns, we cannot simply convert the table to the canonical format, because the data dictionary cache and all possibly existing references to it from other client connection threads would have to be adjusted. row_quiesce_write_index_fields(): Do not crash when the table contains an instantly dropped column. Thanks to Thirunarayanan Balathandayuthapani for discussing the design and implementing an initial prototype of this. Thanks to Matthias Leich for testing.
7 years ago
MDEV-15662 Instant DROP COLUMN or changing the order of columns Allow ADD COLUMN anywhere in a table, not only adding as the last column. Allow instant DROP COLUMN and instant changing the order of columns. The added columns will always be added last in clustered index records. In new records, instantly dropped columns will be stored as NULL or empty when possible. Information about dropped and reordered columns will be written in a metadata BLOB (mblob), which is stored before the first 'user' field in the hidden metadata record at the start of the clustered index. The presence of mblob is indicated by setting the delete-mark flag in the metadata record. The metadata BLOB stores the number of clustered index fields, followed by an array of column information for each field. For dropped columns, we store the NOT NULL flag, the fixed length, and for variable-length columns, whether the maximum length exceeded 255 bytes. For non-dropped columns, we store the column position. Unlike with MDEV-11369, when a table becomes empty, it cannot be converted back to the canonical format. The reason for this is that other threads may hold cached objects such as row_prebuilt_t::ins_node that could refer to dropped or reordered index fields. For instant DROP COLUMN and ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC, we must store the n_core_null_bytes in the root page, so that the chain of node pointer records can be followed in order to reach the leftmost leaf page where the metadata record is located. If the mblob is present, we will zero-initialize the strings "infimum" and "supremum" in the root page, and use the last byte of "supremum" for storing the number of null bytes (which are allocated but useless on node pointer pages). This is necessary for btr_cur_instant_init_metadata() to be able to navigate to the mblob. If the PRIMARY KEY contains any variable-length column and some nullable columns were instantly dropped, the dict_index_t::n_nullable in the data dictionary could be smaller than it actually is in the non-leaf pages. Because of this, the non-leaf pages could use more bytes for the null flags than the data dictionary expects, and we could be reading the lengths of the variable-length columns from the wrong offset, and thus reading the child page number from wrong place. This is the result of two design mistakes that involve unnecessary storage of data: First, it is nonsense to store any data fields for the leftmost node pointer records, because the comparisons would be resolved by the MIN_REC_FLAG alone. Second, there cannot be any null fields in the clustered index node pointer fields, but we nevertheless reserve space for all the null flags. Limitations (future work): MDEV-17459 Allow instant ALTER TABLE even if FULLTEXT INDEX exists MDEV-17468 Avoid table rebuild on operations on generated columns MDEV-17494 Refuse ALGORITHM=INSTANT when the row size is too large btr_page_reorganize_low(): Preserve any metadata in the root page. Call lock_move_reorganize_page() only after restoring the "infimum" and "supremum" records, to avoid a memcmp() assertion failure. dict_col_t::DROPPED: Magic value for dict_col_t::ind. dict_col_t::clear_instant(): Renamed from dict_col_t::remove_instant(). Do not assert that the column was instantly added, because we sometimes call this unconditionally for all columns. Convert an instantly added column to a "core column". The old name remove_instant() could be mistaken to refer to "instant DROP COLUMN". dict_col_t::is_added(): Rename from dict_col_t::is_instant(). dtype_t::metadata_blob_init(): Initialize the mblob data type. dtuple_t::is_metadata(), dtuple_t::is_alter_metadata(), upd_t::is_metadata(), upd_t::is_alter_metadata(): Check if info_bits refer to a metadata record. dict_table_t::instant: Metadata about dropped or reordered columns. dict_table_t::prepare_instant(): Prepare ha_innobase_inplace_ctx::instant_table for instant ALTER TABLE. innobase_instant_try() will pass this to dict_table_t::instant_column(). On rollback, dict_table_t::rollback_instant() will be called. dict_table_t::instant_column(): Renamed from instant_add_column(). Add the parameter col_map so that columns can be reordered. Copy and adjust v_cols[] as well. dict_table_t::find(): Find an old column based on a new column number. dict_table_t::serialise_columns(), dict_table_t::deserialise_columns(): Convert the mblob. dict_index_t::instant_metadata(): Create the metadata record for instant ALTER TABLE. Invoke dict_table_t::serialise_columns(). dict_index_t::reconstruct_fields(): Invoked by dict_table_t::deserialise_columns(). dict_index_t::clear_instant_alter(): Move the fields for the dropped columns to the end, and sort the surviving index fields in ascending order of column position. ha_innobase::check_if_supported_inplace_alter(): Do not allow adding a FTS_DOC_ID column if a hidden FTS_DOC_ID column exists due to FULLTEXT INDEX. (This always required ALGORITHM=COPY.) instant_alter_column_possible(): Add a parameter for InnoDB table, to check for additional conditions, such as the maximum number of index fields. ha_innobase_inplace_ctx::first_alter_pos: The first column whose position is affected by instant ADD, DROP, or changing the order of columns. innobase_build_col_map(): Skip added virtual columns. prepare_inplace_add_virtual(): Correctly compute num_to_add_vcol. Remove some unnecessary code. Note that the call to innodb_base_col_setup() should be executed later. commit_try_norebuild(): If ctx->is_instant(), let the virtual columns be added or dropped by innobase_instant_try(). innobase_instant_try(): Fill in a zero default value for the hidden column FTS_DOC_ID (to reduce the work needed in MDEV-17459). If any columns were dropped or reordered (or added not last), delete any SYS_COLUMNS records for the following columns, and insert SYS_COLUMNS records for all subsequent stored columns as well as for all virtual columns. If any virtual column is dropped, rewrite all virtual column metadata. Use a shortcut only for adding virtual columns. This is because innobase_drop_virtual_try() assumes that the dropped virtual columns still exist in ctx->old_table. innodb_update_cols(): Renamed from innodb_update_n_cols(). innobase_add_one_virtual(), innobase_insert_sys_virtual(): Change the return type to bool, and invoke my_error() when detecting an error. innodb_insert_sys_columns(): Insert a record into SYS_COLUMNS. Refactored from innobase_add_one_virtual() and innobase_instant_add_col(). innobase_instant_add_col(): Replace the parameter dfield with type. innobase_instant_drop_cols(): Drop matching columns from SYS_COLUMNS and all columns from SYS_VIRTUAL. innobase_add_virtual_try(), innobase_drop_virtual_try(): Let the caller invoke innodb_update_cols(). innobase_rename_column_try(): Skip dropped columns. commit_cache_norebuild(): Update table->fts->doc_col. dict_mem_table_col_rename_low(): Skip dropped columns. trx_undo_rec_get_partial_row(): Skip dropped columns. trx_undo_update_rec_get_update(): Handle the metadata BLOB correctly. trx_undo_page_report_modify(): Avoid out-of-bounds access to record fields. Log metadata records consistently. Apparently, the first fields of a clustered index may be updated in an update_undo vector when the index is ID_IND of SYS_FOREIGN, as part of renaming the table during ALTER TABLE. Normally, updates of the PRIMARY KEY should be logged as delete-mark and an insert. row_undo_mod_parse_undo_rec(), row_purge_parse_undo_rec(): Use trx_undo_metadata. row_undo_mod_clust_low(): On metadata rollback, roll back the root page too. row_undo_mod_clust(): Relax an assertion. The delete-mark flag was repurposed for ALTER TABLE metadata records. row_rec_to_index_entry_impl(): Add the template parameter mblob and the optional parameter info_bits for specifying the desired new info bits. For the metadata tuple, allow conversion between the original format (ADD COLUMN only) and the generic format (with hidden BLOB). Add the optional parameter "pad" to determine whether the tuple should be padded to the index fields (on ALTER TABLE it should), or whether it should remain at its original size (on rollback). row_build_index_entry_low(): Clean up the code, removing redundant variables and conditions. For instantly dropped columns, generate a dummy value that is NULL, the empty string, or a fixed length of NUL bytes, depending on the type of the dropped column. row_upd_clust_rec_by_insert_inherit_func(): On the update of PRIMARY KEY of a record that contained a dropped column whose value was stored externally, we will be inserting a dummy NULL or empty string value to the field of the dropped column. The externally stored column would eventually be dropped when purge removes the delete-marked record for the old PRIMARY KEY value. btr_index_rec_validate(): Recognize the metadata record. btr_discard_only_page_on_level(): Preserve the generic instant ALTER TABLE metadata. btr_set_instant(): Replaces page_set_instant(). This sets a clustered index root page to the appropriate format, or upgrades from the MDEV-11369 instant ADD COLUMN to generic ALTER TABLE format. btr_cur_instant_init_low(): Read and validate the metadata BLOB page before reconstructing the dictionary information based on it. btr_cur_instant_init_metadata(): Do not read any lengths from the metadata record header before reading the BLOB. At this point, we would not actually know how many nullable fields the metadata record contains. btr_cur_instant_root_init(): Initialize n_core_null_bytes in one of two possible ways. btr_cur_trim(): Handle the mblob record. row_metadata_to_tuple(): Convert a metadata record to a data tuple, based on the new info_bits of the metadata record. btr_cur_pessimistic_update(): Invoke row_metadata_to_tuple() if needed. Invoke dtuple_convert_big_rec() for metadata records if the record is too large, or if the mblob is not yet marked as externally stored. btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete(): When the last user record is deleted, do not delete the generic instant ALTER TABLE metadata record. Only delete MDEV-11369 instant ADD COLUMN metadata records. btr_cur_optimistic_insert(): Avoid unnecessary computation of rec_size. btr_pcur_store_position(): Allow a logically empty page to contain a metadata record for generic ALTER TABLE. REC_INFO_DEFAULT_ROW_ADD: Renamed from REC_INFO_DEFAULT_ROW. This is for the old instant ADD COLUMN (MDEV-11369) only. REC_INFO_DEFAULT_ROW_ALTER: The more generic metadata record, with additional information for dropped or reordered columns. rec_info_bits_valid(): Remove. The only case when this would fail is when the record is the generic ALTER TABLE metadata record. rec_is_alter_metadata(): Check if a record is the metadata record for instant ALTER TABLE (other than ADD COLUMN). NOTE: This function must not be invoked on node pointer records, because the delete-mark flag in those records may be set (it is garbage), and then a debug assertion could fail because index->is_instant() does not necessarily hold. rec_is_add_metadata(): Check if a record is MDEV-11369 ADD COLUMN metadata record (not more generic instant ALTER TABLE). rec_get_converted_size_comp_prefix_low(): Assume that the metadata field will be stored externally. In dtuple_convert_big_rec() during the rec_get_converted_size() call, it would not be there yet. rec_get_converted_size_comp(): Replace status,fields,n_fields with tuple. rec_init_offsets_comp_ordinary(), rec_get_converted_size_comp_prefix_low(), rec_convert_dtuple_to_rec_comp(): Add template<bool mblob = false>. With mblob=true, process a record with a metadata BLOB. rec_copy_prefix_to_buf(): Assert that no fields beyond the key and system columns are being copied. Exclude the metadata BLOB field. rec_convert_dtuple_to_metadata_comp(): Convert an alter metadata tuple into a record. row_upd_index_replace_metadata(): Apply an update vector to an alter_metadata tuple. row_log_allocate(): Replace dict_index_t::is_instant() with a more appropriate condition that ignores dict_table_t::instant. Only a table on which the MDEV-11369 ADD COLUMN was performed can "lose its instantness" when it becomes empty. After instant DROP COLUMN or reordering columns, we cannot simply convert the table to the canonical format, because the data dictionary cache and all possibly existing references to it from other client connection threads would have to be adjusted. row_quiesce_write_index_fields(): Do not crash when the table contains an instantly dropped column. Thanks to Thirunarayanan Balathandayuthapani for discussing the design and implementing an initial prototype of this. Thanks to Matthias Leich for testing.
7 years ago
MDEV-15662 Instant DROP COLUMN or changing the order of columns Allow ADD COLUMN anywhere in a table, not only adding as the last column. Allow instant DROP COLUMN and instant changing the order of columns. The added columns will always be added last in clustered index records. In new records, instantly dropped columns will be stored as NULL or empty when possible. Information about dropped and reordered columns will be written in a metadata BLOB (mblob), which is stored before the first 'user' field in the hidden metadata record at the start of the clustered index. The presence of mblob is indicated by setting the delete-mark flag in the metadata record. The metadata BLOB stores the number of clustered index fields, followed by an array of column information for each field. For dropped columns, we store the NOT NULL flag, the fixed length, and for variable-length columns, whether the maximum length exceeded 255 bytes. For non-dropped columns, we store the column position. Unlike with MDEV-11369, when a table becomes empty, it cannot be converted back to the canonical format. The reason for this is that other threads may hold cached objects such as row_prebuilt_t::ins_node that could refer to dropped or reordered index fields. For instant DROP COLUMN and ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC, we must store the n_core_null_bytes in the root page, so that the chain of node pointer records can be followed in order to reach the leftmost leaf page where the metadata record is located. If the mblob is present, we will zero-initialize the strings "infimum" and "supremum" in the root page, and use the last byte of "supremum" for storing the number of null bytes (which are allocated but useless on node pointer pages). This is necessary for btr_cur_instant_init_metadata() to be able to navigate to the mblob. If the PRIMARY KEY contains any variable-length column and some nullable columns were instantly dropped, the dict_index_t::n_nullable in the data dictionary could be smaller than it actually is in the non-leaf pages. Because of this, the non-leaf pages could use more bytes for the null flags than the data dictionary expects, and we could be reading the lengths of the variable-length columns from the wrong offset, and thus reading the child page number from wrong place. This is the result of two design mistakes that involve unnecessary storage of data: First, it is nonsense to store any data fields for the leftmost node pointer records, because the comparisons would be resolved by the MIN_REC_FLAG alone. Second, there cannot be any null fields in the clustered index node pointer fields, but we nevertheless reserve space for all the null flags. Limitations (future work): MDEV-17459 Allow instant ALTER TABLE even if FULLTEXT INDEX exists MDEV-17468 Avoid table rebuild on operations on generated columns MDEV-17494 Refuse ALGORITHM=INSTANT when the row size is too large btr_page_reorganize_low(): Preserve any metadata in the root page. Call lock_move_reorganize_page() only after restoring the "infimum" and "supremum" records, to avoid a memcmp() assertion failure. dict_col_t::DROPPED: Magic value for dict_col_t::ind. dict_col_t::clear_instant(): Renamed from dict_col_t::remove_instant(). Do not assert that the column was instantly added, because we sometimes call this unconditionally for all columns. Convert an instantly added column to a "core column". The old name remove_instant() could be mistaken to refer to "instant DROP COLUMN". dict_col_t::is_added(): Rename from dict_col_t::is_instant(). dtype_t::metadata_blob_init(): Initialize the mblob data type. dtuple_t::is_metadata(), dtuple_t::is_alter_metadata(), upd_t::is_metadata(), upd_t::is_alter_metadata(): Check if info_bits refer to a metadata record. dict_table_t::instant: Metadata about dropped or reordered columns. dict_table_t::prepare_instant(): Prepare ha_innobase_inplace_ctx::instant_table for instant ALTER TABLE. innobase_instant_try() will pass this to dict_table_t::instant_column(). On rollback, dict_table_t::rollback_instant() will be called. dict_table_t::instant_column(): Renamed from instant_add_column(). Add the parameter col_map so that columns can be reordered. Copy and adjust v_cols[] as well. dict_table_t::find(): Find an old column based on a new column number. dict_table_t::serialise_columns(), dict_table_t::deserialise_columns(): Convert the mblob. dict_index_t::instant_metadata(): Create the metadata record for instant ALTER TABLE. Invoke dict_table_t::serialise_columns(). dict_index_t::reconstruct_fields(): Invoked by dict_table_t::deserialise_columns(). dict_index_t::clear_instant_alter(): Move the fields for the dropped columns to the end, and sort the surviving index fields in ascending order of column position. ha_innobase::check_if_supported_inplace_alter(): Do not allow adding a FTS_DOC_ID column if a hidden FTS_DOC_ID column exists due to FULLTEXT INDEX. (This always required ALGORITHM=COPY.) instant_alter_column_possible(): Add a parameter for InnoDB table, to check for additional conditions, such as the maximum number of index fields. ha_innobase_inplace_ctx::first_alter_pos: The first column whose position is affected by instant ADD, DROP, or changing the order of columns. innobase_build_col_map(): Skip added virtual columns. prepare_inplace_add_virtual(): Correctly compute num_to_add_vcol. Remove some unnecessary code. Note that the call to innodb_base_col_setup() should be executed later. commit_try_norebuild(): If ctx->is_instant(), let the virtual columns be added or dropped by innobase_instant_try(). innobase_instant_try(): Fill in a zero default value for the hidden column FTS_DOC_ID (to reduce the work needed in MDEV-17459). If any columns were dropped or reordered (or added not last), delete any SYS_COLUMNS records for the following columns, and insert SYS_COLUMNS records for all subsequent stored columns as well as for all virtual columns. If any virtual column is dropped, rewrite all virtual column metadata. Use a shortcut only for adding virtual columns. This is because innobase_drop_virtual_try() assumes that the dropped virtual columns still exist in ctx->old_table. innodb_update_cols(): Renamed from innodb_update_n_cols(). innobase_add_one_virtual(), innobase_insert_sys_virtual(): Change the return type to bool, and invoke my_error() when detecting an error. innodb_insert_sys_columns(): Insert a record into SYS_COLUMNS. Refactored from innobase_add_one_virtual() and innobase_instant_add_col(). innobase_instant_add_col(): Replace the parameter dfield with type. innobase_instant_drop_cols(): Drop matching columns from SYS_COLUMNS and all columns from SYS_VIRTUAL. innobase_add_virtual_try(), innobase_drop_virtual_try(): Let the caller invoke innodb_update_cols(). innobase_rename_column_try(): Skip dropped columns. commit_cache_norebuild(): Update table->fts->doc_col. dict_mem_table_col_rename_low(): Skip dropped columns. trx_undo_rec_get_partial_row(): Skip dropped columns. trx_undo_update_rec_get_update(): Handle the metadata BLOB correctly. trx_undo_page_report_modify(): Avoid out-of-bounds access to record fields. Log metadata records consistently. Apparently, the first fields of a clustered index may be updated in an update_undo vector when the index is ID_IND of SYS_FOREIGN, as part of renaming the table during ALTER TABLE. Normally, updates of the PRIMARY KEY should be logged as delete-mark and an insert. row_undo_mod_parse_undo_rec(), row_purge_parse_undo_rec(): Use trx_undo_metadata. row_undo_mod_clust_low(): On metadata rollback, roll back the root page too. row_undo_mod_clust(): Relax an assertion. The delete-mark flag was repurposed for ALTER TABLE metadata records. row_rec_to_index_entry_impl(): Add the template parameter mblob and the optional parameter info_bits for specifying the desired new info bits. For the metadata tuple, allow conversion between the original format (ADD COLUMN only) and the generic format (with hidden BLOB). Add the optional parameter "pad" to determine whether the tuple should be padded to the index fields (on ALTER TABLE it should), or whether it should remain at its original size (on rollback). row_build_index_entry_low(): Clean up the code, removing redundant variables and conditions. For instantly dropped columns, generate a dummy value that is NULL, the empty string, or a fixed length of NUL bytes, depending on the type of the dropped column. row_upd_clust_rec_by_insert_inherit_func(): On the update of PRIMARY KEY of a record that contained a dropped column whose value was stored externally, we will be inserting a dummy NULL or empty string value to the field of the dropped column. The externally stored column would eventually be dropped when purge removes the delete-marked record for the old PRIMARY KEY value. btr_index_rec_validate(): Recognize the metadata record. btr_discard_only_page_on_level(): Preserve the generic instant ALTER TABLE metadata. btr_set_instant(): Replaces page_set_instant(). This sets a clustered index root page to the appropriate format, or upgrades from the MDEV-11369 instant ADD COLUMN to generic ALTER TABLE format. btr_cur_instant_init_low(): Read and validate the metadata BLOB page before reconstructing the dictionary information based on it. btr_cur_instant_init_metadata(): Do not read any lengths from the metadata record header before reading the BLOB. At this point, we would not actually know how many nullable fields the metadata record contains. btr_cur_instant_root_init(): Initialize n_core_null_bytes in one of two possible ways. btr_cur_trim(): Handle the mblob record. row_metadata_to_tuple(): Convert a metadata record to a data tuple, based on the new info_bits of the metadata record. btr_cur_pessimistic_update(): Invoke row_metadata_to_tuple() if needed. Invoke dtuple_convert_big_rec() for metadata records if the record is too large, or if the mblob is not yet marked as externally stored. btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete(): When the last user record is deleted, do not delete the generic instant ALTER TABLE metadata record. Only delete MDEV-11369 instant ADD COLUMN metadata records. btr_cur_optimistic_insert(): Avoid unnecessary computation of rec_size. btr_pcur_store_position(): Allow a logically empty page to contain a metadata record for generic ALTER TABLE. REC_INFO_DEFAULT_ROW_ADD: Renamed from REC_INFO_DEFAULT_ROW. This is for the old instant ADD COLUMN (MDEV-11369) only. REC_INFO_DEFAULT_ROW_ALTER: The more generic metadata record, with additional information for dropped or reordered columns. rec_info_bits_valid(): Remove. The only case when this would fail is when the record is the generic ALTER TABLE metadata record. rec_is_alter_metadata(): Check if a record is the metadata record for instant ALTER TABLE (other than ADD COLUMN). NOTE: This function must not be invoked on node pointer records, because the delete-mark flag in those records may be set (it is garbage), and then a debug assertion could fail because index->is_instant() does not necessarily hold. rec_is_add_metadata(): Check if a record is MDEV-11369 ADD COLUMN metadata record (not more generic instant ALTER TABLE). rec_get_converted_size_comp_prefix_low(): Assume that the metadata field will be stored externally. In dtuple_convert_big_rec() during the rec_get_converted_size() call, it would not be there yet. rec_get_converted_size_comp(): Replace status,fields,n_fields with tuple. rec_init_offsets_comp_ordinary(), rec_get_converted_size_comp_prefix_low(), rec_convert_dtuple_to_rec_comp(): Add template<bool mblob = false>. With mblob=true, process a record with a metadata BLOB. rec_copy_prefix_to_buf(): Assert that no fields beyond the key and system columns are being copied. Exclude the metadata BLOB field. rec_convert_dtuple_to_metadata_comp(): Convert an alter metadata tuple into a record. row_upd_index_replace_metadata(): Apply an update vector to an alter_metadata tuple. row_log_allocate(): Replace dict_index_t::is_instant() with a more appropriate condition that ignores dict_table_t::instant. Only a table on which the MDEV-11369 ADD COLUMN was performed can "lose its instantness" when it becomes empty. After instant DROP COLUMN or reordering columns, we cannot simply convert the table to the canonical format, because the data dictionary cache and all possibly existing references to it from other client connection threads would have to be adjusted. row_quiesce_write_index_fields(): Do not crash when the table contains an instantly dropped column. Thanks to Thirunarayanan Balathandayuthapani for discussing the design and implementing an initial prototype of this. Thanks to Matthias Leich for testing.
7 years ago
MDEV-11369 Instant ADD COLUMN for InnoDB For InnoDB tables, adding, dropping and reordering columns has required a rebuild of the table and all its indexes. Since MySQL 5.6 (and MariaDB 10.0) this has been supported online (LOCK=NONE), allowing concurrent modification of the tables. This work revises the InnoDB ROW_FORMAT=REDUNDANT, ROW_FORMAT=COMPACT and ROW_FORMAT=DYNAMIC so that columns can be appended instantaneously, with only minor changes performed to the table structure. The counter innodb_instant_alter_column in INFORMATION_SCHEMA.GLOBAL_STATUS is incremented whenever a table rebuild operation is converted into an instant ADD COLUMN operation. ROW_FORMAT=COMPRESSED tables will not support instant ADD COLUMN. Some usability limitations will be addressed in subsequent work: MDEV-13134 Introduce ALTER TABLE attributes ALGORITHM=NOCOPY and ALGORITHM=INSTANT MDEV-14016 Allow instant ADD COLUMN, ADD INDEX, LOCK=NONE The format of the clustered index (PRIMARY KEY) is changed as follows: (1) The FIL_PAGE_TYPE of the root page will be FIL_PAGE_TYPE_INSTANT, and a new field PAGE_INSTANT will contain the original number of fields in the clustered index ('core' fields). If instant ADD COLUMN has not been used or the table becomes empty, or the very first instant ADD COLUMN operation is rolled back, the fields PAGE_INSTANT and FIL_PAGE_TYPE will be reset to 0 and FIL_PAGE_INDEX. (2) A special 'default row' record is inserted into the leftmost leaf, between the page infimum and the first user record. This record is distinguished by the REC_INFO_MIN_REC_FLAG, and it is otherwise in the same format as records that contain values for the instantly added columns. This 'default row' always has the same number of fields as the clustered index according to the table definition. The values of 'core' fields are to be ignored. For other fields, the 'default row' will contain the default values as they were during the ALTER TABLE statement. (If the column default values are changed later, those values will only be stored in the .frm file. The 'default row' will contain the original evaluated values, which must be the same for every row.) The 'default row' must be completely hidden from higher-level access routines. Assertions have been added to ensure that no 'default row' is ever present in the adaptive hash index or in locked records. The 'default row' is never delete-marked. (3) In clustered index leaf page records, the number of fields must reside between the number of 'core' fields (dict_index_t::n_core_fields introduced in this work) and dict_index_t::n_fields. If the number of fields is less than dict_index_t::n_fields, the missing fields are replaced with the column value of the 'default row'. Note: The number of fields in the record may shrink if some of the last instantly added columns are updated to the value that is in the 'default row'. The function btr_cur_trim() implements this 'compression' on update and rollback; dtuple::trim() implements it on insert. (4) In ROW_FORMAT=COMPACT and ROW_FORMAT=DYNAMIC records, the new status value REC_STATUS_COLUMNS_ADDED will indicate the presence of a new record header that will encode n_fields-n_core_fields-1 in 1 or 2 bytes. (In ROW_FORMAT=REDUNDANT records, the record header always explicitly encodes the number of fields.) We introduce the undo log record type TRX_UNDO_INSERT_DEFAULT for covering the insert of the 'default row' record when instant ADD COLUMN is used for the first time. Subsequent instant ADD COLUMN can use TRX_UNDO_UPD_EXIST_REC. This is joint work with Vin Chen (陈福荣) from Tencent. The design that was discussed in April 2017 would not have allowed import or export of data files, because instead of the 'default row' it would have introduced a data dictionary table. The test rpl.rpl_alter_instant is exactly as contributed in pull request #408. The test innodb.instant_alter is based on a contributed test. The redo log record format changes for ROW_FORMAT=DYNAMIC and ROW_FORMAT=COMPACT are as contributed. (With this change present, crash recovery from MariaDB 10.3.1 will fail in spectacular ways!) Also the semantics of higher-level redo log records that modify the PAGE_INSTANT field is changed. The redo log format version identifier was already changed to LOG_HEADER_FORMAT_CURRENT=103 in MariaDB 10.3.1. Everything else has been rewritten by me. Thanks to Elena Stepanova, the code has been tested extensively. When rolling back an instant ADD COLUMN operation, we must empty the PAGE_FREE list after deleting or shortening the 'default row' record, by calling either btr_page_empty() or btr_page_reorganize(). We must know the size of each entry in the PAGE_FREE list. If rollback left a freed copy of the 'default row' in the PAGE_FREE list, we would be unable to determine its size (if it is in ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC) because it would contain more fields than the rolled-back definition of the clustered index. UNIV_SQL_DEFAULT: A new special constant that designates an instantly added column that is not present in the clustered index record. len_is_stored(): Check if a length is an actual length. There are two magic length values: UNIV_SQL_DEFAULT, UNIV_SQL_NULL. dict_col_t::def_val: The 'default row' value of the column. If the column is not added instantly, def_val.len will be UNIV_SQL_DEFAULT. dict_col_t: Add the accessors is_virtual(), is_nullable(), is_instant(), instant_value(). dict_col_t::remove_instant(): Remove the 'instant ADD' status of a column. dict_col_t::name(const dict_table_t& table): Replaces dict_table_get_col_name(). dict_index_t::n_core_fields: The original number of fields. For secondary indexes and if instant ADD COLUMN has not been used, this will be equal to dict_index_t::n_fields. dict_index_t::n_core_null_bytes: Number of bytes needed to represent the null flags; usually equal to UT_BITS_IN_BYTES(n_nullable). dict_index_t::NO_CORE_NULL_BYTES: Magic value signalling that n_core_null_bytes was not initialized yet from the clustered index root page. dict_index_t: Add the accessors is_instant(), is_clust(), get_n_nullable(), instant_field_value(). dict_index_t::instant_add_field(): Adjust clustered index metadata for instant ADD COLUMN. dict_index_t::remove_instant(): Remove the 'instant ADD' status of a clustered index when the table becomes empty, or the very first instant ADD COLUMN operation is rolled back. dict_table_t: Add the accessors is_instant(), is_temporary(), supports_instant(). dict_table_t::instant_add_column(): Adjust metadata for instant ADD COLUMN. dict_table_t::rollback_instant(): Adjust metadata on the rollback of instant ADD COLUMN. prepare_inplace_alter_table_dict(): First create the ctx->new_table, and only then decide if the table really needs to be rebuilt. We must split the creation of table or index metadata from the creation of the dictionary table records and the creation of the data. In this way, we can transform a table-rebuilding operation into an instant ADD COLUMN operation. Dictionary objects will only be added to cache when table rebuilding or index creation is needed. The ctx->instant_table will never be added to cache. dict_table_t::add_to_cache(): Modified and renamed from dict_table_add_to_cache(). Do not modify the table metadata. Let the callers invoke dict_table_add_system_columns() and if needed, set can_be_evicted. dict_create_sys_tables_tuple(), dict_create_table_step(): Omit the system columns (which will now exist in the dict_table_t object already at this point). dict_create_table_step(): Expect the callers to invoke dict_table_add_system_columns(). pars_create_table(): Before creating the table creation execution graph, invoke dict_table_add_system_columns(). row_create_table_for_mysql(): Expect all callers to invoke dict_table_add_system_columns(). create_index_dict(): Replaces row_merge_create_index_graph(). innodb_update_n_cols(): Renamed from innobase_update_n_virtual(). Call my_error() if an error occurs. btr_cur_instant_init(), btr_cur_instant_init_low(), btr_cur_instant_root_init(): Load additional metadata from the clustered index and set dict_index_t::n_core_null_bytes. This is invoked when table metadata is first loaded into the data dictionary. dict_boot(): Initialize n_core_null_bytes for the four hard-coded dictionary tables. dict_create_index_step(): Initialize n_core_null_bytes. This is executed as part of CREATE TABLE. dict_index_build_internal_clust(): Initialize n_core_null_bytes to NO_CORE_NULL_BYTES if table->supports_instant(). row_create_index_for_mysql(): Initialize n_core_null_bytes for CREATE TEMPORARY TABLE. commit_cache_norebuild(): Call the code to rename or enlarge columns in the cache only if instant ADD COLUMN is not being used. (Instant ADD COLUMN would copy all column metadata from instant_table to old_table, including the names and lengths.) PAGE_INSTANT: A new 13-bit field for storing dict_index_t::n_core_fields. This is repurposing the 16-bit field PAGE_DIRECTION, of which only the least significant 3 bits were used. The original byte containing PAGE_DIRECTION will be accessible via the new constant PAGE_DIRECTION_B. page_get_instant(), page_set_instant(): Accessors for the PAGE_INSTANT. page_ptr_get_direction(), page_get_direction(), page_ptr_set_direction(): Accessors for PAGE_DIRECTION. page_direction_reset(): Reset PAGE_DIRECTION, PAGE_N_DIRECTION. page_direction_increment(): Increment PAGE_N_DIRECTION and set PAGE_DIRECTION. rec_get_offsets(): Use the 'leaf' parameter for non-debug purposes, and assume that heap_no is always set. Initialize all dict_index_t::n_fields for ROW_FORMAT=REDUNDANT records, even if the record contains fewer fields. rec_offs_make_valid(): Add the parameter 'leaf'. rec_copy_prefix_to_dtuple(): Assert that the tuple is only built on the core fields. Instant ADD COLUMN only applies to the clustered index, and we should never build a search key that has more than the PRIMARY KEY and possibly DB_TRX_ID,DB_ROLL_PTR. All these columns are always present. dict_index_build_data_tuple(): Remove assertions that would be duplicated in rec_copy_prefix_to_dtuple(). rec_init_offsets(): Support ROW_FORMAT=REDUNDANT records whose number of fields is between n_core_fields and n_fields. cmp_rec_rec_with_match(): Implement the comparison between two MIN_REC_FLAG records. trx_t::in_rollback: Make the field available in non-debug builds. trx_start_for_ddl_low(): Remove dangerous error-tolerance. A dictionary transaction must be flagged as such before it has generated any undo log records. This is because trx_undo_assign_undo() will mark the transaction as a dictionary transaction in the undo log header right before the very first undo log record is being written. btr_index_rec_validate(): Account for instant ADD COLUMN row_undo_ins_remove_clust_rec(): On the rollback of an insert into SYS_COLUMNS, revert instant ADD COLUMN in the cache by removing the last column from the table and the clustered index. row_search_on_row_ref(), row_undo_mod_parse_undo_rec(), row_undo_mod(), trx_undo_update_rec_get_update(): Handle the 'default row' as a special case. dtuple_t::trim(index): Omit a redundant suffix of an index tuple right before insert or update. After instant ADD COLUMN, if the last fields of a clustered index tuple match the 'default row', there is no need to store them. While trimming the entry, we must hold a page latch, so that the table cannot be emptied and the 'default row' be deleted. btr_cur_optimistic_update(), btr_cur_pessimistic_update(), row_upd_clust_rec_by_insert(), row_ins_clust_index_entry_low(): Invoke dtuple_t::trim() if needed. row_ins_clust_index_entry(): Restore dtuple_t::n_fields after calling row_ins_clust_index_entry_low(). rec_get_converted_size(), rec_get_converted_size_comp(): Allow the number of fields to be between n_core_fields and n_fields. Do not support infimum,supremum. They are never supposed to be stored in dtuple_t, because page creation nowadays uses a lower-level method for initializing them. rec_convert_dtuple_to_rec_comp(): Assign the status bits based on the number of fields. btr_cur_trim(): In an update, trim the index entry as needed. For the 'default row', handle rollback specially. For user records, omit fields that match the 'default row'. btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete(): Skip locking and adaptive hash index for the 'default row'. row_log_table_apply_convert_mrec(): Replace 'default row' values if needed. In the temporary file that is applied by row_log_table_apply(), we must identify whether the records contain the extra header for instantly added columns. For now, we will allocate an additional byte for this for ROW_T_INSERT and ROW_T_UPDATE records when the source table has been subject to instant ADD COLUMN. The ROW_T_DELETE records are fine, as they will be converted and will only contain 'core' columns (PRIMARY KEY and some system columns) that are converted from dtuple_t. rec_get_converted_size_temp(), rec_init_offsets_temp(), rec_convert_dtuple_to_temp(): Add the parameter 'status'. REC_INFO_DEFAULT_ROW = REC_INFO_MIN_REC_FLAG | REC_STATUS_COLUMNS_ADDED: An info_bits constant for distinguishing the 'default row' record. rec_comp_status_t: An enum of the status bit values. rec_leaf_format: An enum that replaces the bool parameter of rec_init_offsets_comp_ordinary().
8 years ago
MDEV-15662 Instant DROP COLUMN or changing the order of columns Allow ADD COLUMN anywhere in a table, not only adding as the last column. Allow instant DROP COLUMN and instant changing the order of columns. The added columns will always be added last in clustered index records. In new records, instantly dropped columns will be stored as NULL or empty when possible. Information about dropped and reordered columns will be written in a metadata BLOB (mblob), which is stored before the first 'user' field in the hidden metadata record at the start of the clustered index. The presence of mblob is indicated by setting the delete-mark flag in the metadata record. The metadata BLOB stores the number of clustered index fields, followed by an array of column information for each field. For dropped columns, we store the NOT NULL flag, the fixed length, and for variable-length columns, whether the maximum length exceeded 255 bytes. For non-dropped columns, we store the column position. Unlike with MDEV-11369, when a table becomes empty, it cannot be converted back to the canonical format. The reason for this is that other threads may hold cached objects such as row_prebuilt_t::ins_node that could refer to dropped or reordered index fields. For instant DROP COLUMN and ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC, we must store the n_core_null_bytes in the root page, so that the chain of node pointer records can be followed in order to reach the leftmost leaf page where the metadata record is located. If the mblob is present, we will zero-initialize the strings "infimum" and "supremum" in the root page, and use the last byte of "supremum" for storing the number of null bytes (which are allocated but useless on node pointer pages). This is necessary for btr_cur_instant_init_metadata() to be able to navigate to the mblob. If the PRIMARY KEY contains any variable-length column and some nullable columns were instantly dropped, the dict_index_t::n_nullable in the data dictionary could be smaller than it actually is in the non-leaf pages. Because of this, the non-leaf pages could use more bytes for the null flags than the data dictionary expects, and we could be reading the lengths of the variable-length columns from the wrong offset, and thus reading the child page number from wrong place. This is the result of two design mistakes that involve unnecessary storage of data: First, it is nonsense to store any data fields for the leftmost node pointer records, because the comparisons would be resolved by the MIN_REC_FLAG alone. Second, there cannot be any null fields in the clustered index node pointer fields, but we nevertheless reserve space for all the null flags. Limitations (future work): MDEV-17459 Allow instant ALTER TABLE even if FULLTEXT INDEX exists MDEV-17468 Avoid table rebuild on operations on generated columns MDEV-17494 Refuse ALGORITHM=INSTANT when the row size is too large btr_page_reorganize_low(): Preserve any metadata in the root page. Call lock_move_reorganize_page() only after restoring the "infimum" and "supremum" records, to avoid a memcmp() assertion failure. dict_col_t::DROPPED: Magic value for dict_col_t::ind. dict_col_t::clear_instant(): Renamed from dict_col_t::remove_instant(). Do not assert that the column was instantly added, because we sometimes call this unconditionally for all columns. Convert an instantly added column to a "core column". The old name remove_instant() could be mistaken to refer to "instant DROP COLUMN". dict_col_t::is_added(): Rename from dict_col_t::is_instant(). dtype_t::metadata_blob_init(): Initialize the mblob data type. dtuple_t::is_metadata(), dtuple_t::is_alter_metadata(), upd_t::is_metadata(), upd_t::is_alter_metadata(): Check if info_bits refer to a metadata record. dict_table_t::instant: Metadata about dropped or reordered columns. dict_table_t::prepare_instant(): Prepare ha_innobase_inplace_ctx::instant_table for instant ALTER TABLE. innobase_instant_try() will pass this to dict_table_t::instant_column(). On rollback, dict_table_t::rollback_instant() will be called. dict_table_t::instant_column(): Renamed from instant_add_column(). Add the parameter col_map so that columns can be reordered. Copy and adjust v_cols[] as well. dict_table_t::find(): Find an old column based on a new column number. dict_table_t::serialise_columns(), dict_table_t::deserialise_columns(): Convert the mblob. dict_index_t::instant_metadata(): Create the metadata record for instant ALTER TABLE. Invoke dict_table_t::serialise_columns(). dict_index_t::reconstruct_fields(): Invoked by dict_table_t::deserialise_columns(). dict_index_t::clear_instant_alter(): Move the fields for the dropped columns to the end, and sort the surviving index fields in ascending order of column position. ha_innobase::check_if_supported_inplace_alter(): Do not allow adding a FTS_DOC_ID column if a hidden FTS_DOC_ID column exists due to FULLTEXT INDEX. (This always required ALGORITHM=COPY.) instant_alter_column_possible(): Add a parameter for InnoDB table, to check for additional conditions, such as the maximum number of index fields. ha_innobase_inplace_ctx::first_alter_pos: The first column whose position is affected by instant ADD, DROP, or changing the order of columns. innobase_build_col_map(): Skip added virtual columns. prepare_inplace_add_virtual(): Correctly compute num_to_add_vcol. Remove some unnecessary code. Note that the call to innodb_base_col_setup() should be executed later. commit_try_norebuild(): If ctx->is_instant(), let the virtual columns be added or dropped by innobase_instant_try(). innobase_instant_try(): Fill in a zero default value for the hidden column FTS_DOC_ID (to reduce the work needed in MDEV-17459). If any columns were dropped or reordered (or added not last), delete any SYS_COLUMNS records for the following columns, and insert SYS_COLUMNS records for all subsequent stored columns as well as for all virtual columns. If any virtual column is dropped, rewrite all virtual column metadata. Use a shortcut only for adding virtual columns. This is because innobase_drop_virtual_try() assumes that the dropped virtual columns still exist in ctx->old_table. innodb_update_cols(): Renamed from innodb_update_n_cols(). innobase_add_one_virtual(), innobase_insert_sys_virtual(): Change the return type to bool, and invoke my_error() when detecting an error. innodb_insert_sys_columns(): Insert a record into SYS_COLUMNS. Refactored from innobase_add_one_virtual() and innobase_instant_add_col(). innobase_instant_add_col(): Replace the parameter dfield with type. innobase_instant_drop_cols(): Drop matching columns from SYS_COLUMNS and all columns from SYS_VIRTUAL. innobase_add_virtual_try(), innobase_drop_virtual_try(): Let the caller invoke innodb_update_cols(). innobase_rename_column_try(): Skip dropped columns. commit_cache_norebuild(): Update table->fts->doc_col. dict_mem_table_col_rename_low(): Skip dropped columns. trx_undo_rec_get_partial_row(): Skip dropped columns. trx_undo_update_rec_get_update(): Handle the metadata BLOB correctly. trx_undo_page_report_modify(): Avoid out-of-bounds access to record fields. Log metadata records consistently. Apparently, the first fields of a clustered index may be updated in an update_undo vector when the index is ID_IND of SYS_FOREIGN, as part of renaming the table during ALTER TABLE. Normally, updates of the PRIMARY KEY should be logged as delete-mark and an insert. row_undo_mod_parse_undo_rec(), row_purge_parse_undo_rec(): Use trx_undo_metadata. row_undo_mod_clust_low(): On metadata rollback, roll back the root page too. row_undo_mod_clust(): Relax an assertion. The delete-mark flag was repurposed for ALTER TABLE metadata records. row_rec_to_index_entry_impl(): Add the template parameter mblob and the optional parameter info_bits for specifying the desired new info bits. For the metadata tuple, allow conversion between the original format (ADD COLUMN only) and the generic format (with hidden BLOB). Add the optional parameter "pad" to determine whether the tuple should be padded to the index fields (on ALTER TABLE it should), or whether it should remain at its original size (on rollback). row_build_index_entry_low(): Clean up the code, removing redundant variables and conditions. For instantly dropped columns, generate a dummy value that is NULL, the empty string, or a fixed length of NUL bytes, depending on the type of the dropped column. row_upd_clust_rec_by_insert_inherit_func(): On the update of PRIMARY KEY of a record that contained a dropped column whose value was stored externally, we will be inserting a dummy NULL or empty string value to the field of the dropped column. The externally stored column would eventually be dropped when purge removes the delete-marked record for the old PRIMARY KEY value. btr_index_rec_validate(): Recognize the metadata record. btr_discard_only_page_on_level(): Preserve the generic instant ALTER TABLE metadata. btr_set_instant(): Replaces page_set_instant(). This sets a clustered index root page to the appropriate format, or upgrades from the MDEV-11369 instant ADD COLUMN to generic ALTER TABLE format. btr_cur_instant_init_low(): Read and validate the metadata BLOB page before reconstructing the dictionary information based on it. btr_cur_instant_init_metadata(): Do not read any lengths from the metadata record header before reading the BLOB. At this point, we would not actually know how many nullable fields the metadata record contains. btr_cur_instant_root_init(): Initialize n_core_null_bytes in one of two possible ways. btr_cur_trim(): Handle the mblob record. row_metadata_to_tuple(): Convert a metadata record to a data tuple, based on the new info_bits of the metadata record. btr_cur_pessimistic_update(): Invoke row_metadata_to_tuple() if needed. Invoke dtuple_convert_big_rec() for metadata records if the record is too large, or if the mblob is not yet marked as externally stored. btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete(): When the last user record is deleted, do not delete the generic instant ALTER TABLE metadata record. Only delete MDEV-11369 instant ADD COLUMN metadata records. btr_cur_optimistic_insert(): Avoid unnecessary computation of rec_size. btr_pcur_store_position(): Allow a logically empty page to contain a metadata record for generic ALTER TABLE. REC_INFO_DEFAULT_ROW_ADD: Renamed from REC_INFO_DEFAULT_ROW. This is for the old instant ADD COLUMN (MDEV-11369) only. REC_INFO_DEFAULT_ROW_ALTER: The more generic metadata record, with additional information for dropped or reordered columns. rec_info_bits_valid(): Remove. The only case when this would fail is when the record is the generic ALTER TABLE metadata record. rec_is_alter_metadata(): Check if a record is the metadata record for instant ALTER TABLE (other than ADD COLUMN). NOTE: This function must not be invoked on node pointer records, because the delete-mark flag in those records may be set (it is garbage), and then a debug assertion could fail because index->is_instant() does not necessarily hold. rec_is_add_metadata(): Check if a record is MDEV-11369 ADD COLUMN metadata record (not more generic instant ALTER TABLE). rec_get_converted_size_comp_prefix_low(): Assume that the metadata field will be stored externally. In dtuple_convert_big_rec() during the rec_get_converted_size() call, it would not be there yet. rec_get_converted_size_comp(): Replace status,fields,n_fields with tuple. rec_init_offsets_comp_ordinary(), rec_get_converted_size_comp_prefix_low(), rec_convert_dtuple_to_rec_comp(): Add template<bool mblob = false>. With mblob=true, process a record with a metadata BLOB. rec_copy_prefix_to_buf(): Assert that no fields beyond the key and system columns are being copied. Exclude the metadata BLOB field. rec_convert_dtuple_to_metadata_comp(): Convert an alter metadata tuple into a record. row_upd_index_replace_metadata(): Apply an update vector to an alter_metadata tuple. row_log_allocate(): Replace dict_index_t::is_instant() with a more appropriate condition that ignores dict_table_t::instant. Only a table on which the MDEV-11369 ADD COLUMN was performed can "lose its instantness" when it becomes empty. After instant DROP COLUMN or reordering columns, we cannot simply convert the table to the canonical format, because the data dictionary cache and all possibly existing references to it from other client connection threads would have to be adjusted. row_quiesce_write_index_fields(): Do not crash when the table contains an instantly dropped column. Thanks to Thirunarayanan Balathandayuthapani for discussing the design and implementing an initial prototype of this. Thanks to Matthias Leich for testing.
7 years ago
MDEV-15662 Instant DROP COLUMN or changing the order of columns Allow ADD COLUMN anywhere in a table, not only adding as the last column. Allow instant DROP COLUMN and instant changing the order of columns. The added columns will always be added last in clustered index records. In new records, instantly dropped columns will be stored as NULL or empty when possible. Information about dropped and reordered columns will be written in a metadata BLOB (mblob), which is stored before the first 'user' field in the hidden metadata record at the start of the clustered index. The presence of mblob is indicated by setting the delete-mark flag in the metadata record. The metadata BLOB stores the number of clustered index fields, followed by an array of column information for each field. For dropped columns, we store the NOT NULL flag, the fixed length, and for variable-length columns, whether the maximum length exceeded 255 bytes. For non-dropped columns, we store the column position. Unlike with MDEV-11369, when a table becomes empty, it cannot be converted back to the canonical format. The reason for this is that other threads may hold cached objects such as row_prebuilt_t::ins_node that could refer to dropped or reordered index fields. For instant DROP COLUMN and ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC, we must store the n_core_null_bytes in the root page, so that the chain of node pointer records can be followed in order to reach the leftmost leaf page where the metadata record is located. If the mblob is present, we will zero-initialize the strings "infimum" and "supremum" in the root page, and use the last byte of "supremum" for storing the number of null bytes (which are allocated but useless on node pointer pages). This is necessary for btr_cur_instant_init_metadata() to be able to navigate to the mblob. If the PRIMARY KEY contains any variable-length column and some nullable columns were instantly dropped, the dict_index_t::n_nullable in the data dictionary could be smaller than it actually is in the non-leaf pages. Because of this, the non-leaf pages could use more bytes for the null flags than the data dictionary expects, and we could be reading the lengths of the variable-length columns from the wrong offset, and thus reading the child page number from wrong place. This is the result of two design mistakes that involve unnecessary storage of data: First, it is nonsense to store any data fields for the leftmost node pointer records, because the comparisons would be resolved by the MIN_REC_FLAG alone. Second, there cannot be any null fields in the clustered index node pointer fields, but we nevertheless reserve space for all the null flags. Limitations (future work): MDEV-17459 Allow instant ALTER TABLE even if FULLTEXT INDEX exists MDEV-17468 Avoid table rebuild on operations on generated columns MDEV-17494 Refuse ALGORITHM=INSTANT when the row size is too large btr_page_reorganize_low(): Preserve any metadata in the root page. Call lock_move_reorganize_page() only after restoring the "infimum" and "supremum" records, to avoid a memcmp() assertion failure. dict_col_t::DROPPED: Magic value for dict_col_t::ind. dict_col_t::clear_instant(): Renamed from dict_col_t::remove_instant(). Do not assert that the column was instantly added, because we sometimes call this unconditionally for all columns. Convert an instantly added column to a "core column". The old name remove_instant() could be mistaken to refer to "instant DROP COLUMN". dict_col_t::is_added(): Rename from dict_col_t::is_instant(). dtype_t::metadata_blob_init(): Initialize the mblob data type. dtuple_t::is_metadata(), dtuple_t::is_alter_metadata(), upd_t::is_metadata(), upd_t::is_alter_metadata(): Check if info_bits refer to a metadata record. dict_table_t::instant: Metadata about dropped or reordered columns. dict_table_t::prepare_instant(): Prepare ha_innobase_inplace_ctx::instant_table for instant ALTER TABLE. innobase_instant_try() will pass this to dict_table_t::instant_column(). On rollback, dict_table_t::rollback_instant() will be called. dict_table_t::instant_column(): Renamed from instant_add_column(). Add the parameter col_map so that columns can be reordered. Copy and adjust v_cols[] as well. dict_table_t::find(): Find an old column based on a new column number. dict_table_t::serialise_columns(), dict_table_t::deserialise_columns(): Convert the mblob. dict_index_t::instant_metadata(): Create the metadata record for instant ALTER TABLE. Invoke dict_table_t::serialise_columns(). dict_index_t::reconstruct_fields(): Invoked by dict_table_t::deserialise_columns(). dict_index_t::clear_instant_alter(): Move the fields for the dropped columns to the end, and sort the surviving index fields in ascending order of column position. ha_innobase::check_if_supported_inplace_alter(): Do not allow adding a FTS_DOC_ID column if a hidden FTS_DOC_ID column exists due to FULLTEXT INDEX. (This always required ALGORITHM=COPY.) instant_alter_column_possible(): Add a parameter for InnoDB table, to check for additional conditions, such as the maximum number of index fields. ha_innobase_inplace_ctx::first_alter_pos: The first column whose position is affected by instant ADD, DROP, or changing the order of columns. innobase_build_col_map(): Skip added virtual columns. prepare_inplace_add_virtual(): Correctly compute num_to_add_vcol. Remove some unnecessary code. Note that the call to innodb_base_col_setup() should be executed later. commit_try_norebuild(): If ctx->is_instant(), let the virtual columns be added or dropped by innobase_instant_try(). innobase_instant_try(): Fill in a zero default value for the hidden column FTS_DOC_ID (to reduce the work needed in MDEV-17459). If any columns were dropped or reordered (or added not last), delete any SYS_COLUMNS records for the following columns, and insert SYS_COLUMNS records for all subsequent stored columns as well as for all virtual columns. If any virtual column is dropped, rewrite all virtual column metadata. Use a shortcut only for adding virtual columns. This is because innobase_drop_virtual_try() assumes that the dropped virtual columns still exist in ctx->old_table. innodb_update_cols(): Renamed from innodb_update_n_cols(). innobase_add_one_virtual(), innobase_insert_sys_virtual(): Change the return type to bool, and invoke my_error() when detecting an error. innodb_insert_sys_columns(): Insert a record into SYS_COLUMNS. Refactored from innobase_add_one_virtual() and innobase_instant_add_col(). innobase_instant_add_col(): Replace the parameter dfield with type. innobase_instant_drop_cols(): Drop matching columns from SYS_COLUMNS and all columns from SYS_VIRTUAL. innobase_add_virtual_try(), innobase_drop_virtual_try(): Let the caller invoke innodb_update_cols(). innobase_rename_column_try(): Skip dropped columns. commit_cache_norebuild(): Update table->fts->doc_col. dict_mem_table_col_rename_low(): Skip dropped columns. trx_undo_rec_get_partial_row(): Skip dropped columns. trx_undo_update_rec_get_update(): Handle the metadata BLOB correctly. trx_undo_page_report_modify(): Avoid out-of-bounds access to record fields. Log metadata records consistently. Apparently, the first fields of a clustered index may be updated in an update_undo vector when the index is ID_IND of SYS_FOREIGN, as part of renaming the table during ALTER TABLE. Normally, updates of the PRIMARY KEY should be logged as delete-mark and an insert. row_undo_mod_parse_undo_rec(), row_purge_parse_undo_rec(): Use trx_undo_metadata. row_undo_mod_clust_low(): On metadata rollback, roll back the root page too. row_undo_mod_clust(): Relax an assertion. The delete-mark flag was repurposed for ALTER TABLE metadata records. row_rec_to_index_entry_impl(): Add the template parameter mblob and the optional parameter info_bits for specifying the desired new info bits. For the metadata tuple, allow conversion between the original format (ADD COLUMN only) and the generic format (with hidden BLOB). Add the optional parameter "pad" to determine whether the tuple should be padded to the index fields (on ALTER TABLE it should), or whether it should remain at its original size (on rollback). row_build_index_entry_low(): Clean up the code, removing redundant variables and conditions. For instantly dropped columns, generate a dummy value that is NULL, the empty string, or a fixed length of NUL bytes, depending on the type of the dropped column. row_upd_clust_rec_by_insert_inherit_func(): On the update of PRIMARY KEY of a record that contained a dropped column whose value was stored externally, we will be inserting a dummy NULL or empty string value to the field of the dropped column. The externally stored column would eventually be dropped when purge removes the delete-marked record for the old PRIMARY KEY value. btr_index_rec_validate(): Recognize the metadata record. btr_discard_only_page_on_level(): Preserve the generic instant ALTER TABLE metadata. btr_set_instant(): Replaces page_set_instant(). This sets a clustered index root page to the appropriate format, or upgrades from the MDEV-11369 instant ADD COLUMN to generic ALTER TABLE format. btr_cur_instant_init_low(): Read and validate the metadata BLOB page before reconstructing the dictionary information based on it. btr_cur_instant_init_metadata(): Do not read any lengths from the metadata record header before reading the BLOB. At this point, we would not actually know how many nullable fields the metadata record contains. btr_cur_instant_root_init(): Initialize n_core_null_bytes in one of two possible ways. btr_cur_trim(): Handle the mblob record. row_metadata_to_tuple(): Convert a metadata record to a data tuple, based on the new info_bits of the metadata record. btr_cur_pessimistic_update(): Invoke row_metadata_to_tuple() if needed. Invoke dtuple_convert_big_rec() for metadata records if the record is too large, or if the mblob is not yet marked as externally stored. btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete(): When the last user record is deleted, do not delete the generic instant ALTER TABLE metadata record. Only delete MDEV-11369 instant ADD COLUMN metadata records. btr_cur_optimistic_insert(): Avoid unnecessary computation of rec_size. btr_pcur_store_position(): Allow a logically empty page to contain a metadata record for generic ALTER TABLE. REC_INFO_DEFAULT_ROW_ADD: Renamed from REC_INFO_DEFAULT_ROW. This is for the old instant ADD COLUMN (MDEV-11369) only. REC_INFO_DEFAULT_ROW_ALTER: The more generic metadata record, with additional information for dropped or reordered columns. rec_info_bits_valid(): Remove. The only case when this would fail is when the record is the generic ALTER TABLE metadata record. rec_is_alter_metadata(): Check if a record is the metadata record for instant ALTER TABLE (other than ADD COLUMN). NOTE: This function must not be invoked on node pointer records, because the delete-mark flag in those records may be set (it is garbage), and then a debug assertion could fail because index->is_instant() does not necessarily hold. rec_is_add_metadata(): Check if a record is MDEV-11369 ADD COLUMN metadata record (not more generic instant ALTER TABLE). rec_get_converted_size_comp_prefix_low(): Assume that the metadata field will be stored externally. In dtuple_convert_big_rec() during the rec_get_converted_size() call, it would not be there yet. rec_get_converted_size_comp(): Replace status,fields,n_fields with tuple. rec_init_offsets_comp_ordinary(), rec_get_converted_size_comp_prefix_low(), rec_convert_dtuple_to_rec_comp(): Add template<bool mblob = false>. With mblob=true, process a record with a metadata BLOB. rec_copy_prefix_to_buf(): Assert that no fields beyond the key and system columns are being copied. Exclude the metadata BLOB field. rec_convert_dtuple_to_metadata_comp(): Convert an alter metadata tuple into a record. row_upd_index_replace_metadata(): Apply an update vector to an alter_metadata tuple. row_log_allocate(): Replace dict_index_t::is_instant() with a more appropriate condition that ignores dict_table_t::instant. Only a table on which the MDEV-11369 ADD COLUMN was performed can "lose its instantness" when it becomes empty. After instant DROP COLUMN or reordering columns, we cannot simply convert the table to the canonical format, because the data dictionary cache and all possibly existing references to it from other client connection threads would have to be adjusted. row_quiesce_write_index_fields(): Do not crash when the table contains an instantly dropped column. Thanks to Thirunarayanan Balathandayuthapani for discussing the design and implementing an initial prototype of this. Thanks to Matthias Leich for testing.
7 years ago
MDEV-11369 Instant ADD COLUMN for InnoDB For InnoDB tables, adding, dropping and reordering columns has required a rebuild of the table and all its indexes. Since MySQL 5.6 (and MariaDB 10.0) this has been supported online (LOCK=NONE), allowing concurrent modification of the tables. This work revises the InnoDB ROW_FORMAT=REDUNDANT, ROW_FORMAT=COMPACT and ROW_FORMAT=DYNAMIC so that columns can be appended instantaneously, with only minor changes performed to the table structure. The counter innodb_instant_alter_column in INFORMATION_SCHEMA.GLOBAL_STATUS is incremented whenever a table rebuild operation is converted into an instant ADD COLUMN operation. ROW_FORMAT=COMPRESSED tables will not support instant ADD COLUMN. Some usability limitations will be addressed in subsequent work: MDEV-13134 Introduce ALTER TABLE attributes ALGORITHM=NOCOPY and ALGORITHM=INSTANT MDEV-14016 Allow instant ADD COLUMN, ADD INDEX, LOCK=NONE The format of the clustered index (PRIMARY KEY) is changed as follows: (1) The FIL_PAGE_TYPE of the root page will be FIL_PAGE_TYPE_INSTANT, and a new field PAGE_INSTANT will contain the original number of fields in the clustered index ('core' fields). If instant ADD COLUMN has not been used or the table becomes empty, or the very first instant ADD COLUMN operation is rolled back, the fields PAGE_INSTANT and FIL_PAGE_TYPE will be reset to 0 and FIL_PAGE_INDEX. (2) A special 'default row' record is inserted into the leftmost leaf, between the page infimum and the first user record. This record is distinguished by the REC_INFO_MIN_REC_FLAG, and it is otherwise in the same format as records that contain values for the instantly added columns. This 'default row' always has the same number of fields as the clustered index according to the table definition. The values of 'core' fields are to be ignored. For other fields, the 'default row' will contain the default values as they were during the ALTER TABLE statement. (If the column default values are changed later, those values will only be stored in the .frm file. The 'default row' will contain the original evaluated values, which must be the same for every row.) The 'default row' must be completely hidden from higher-level access routines. Assertions have been added to ensure that no 'default row' is ever present in the adaptive hash index or in locked records. The 'default row' is never delete-marked. (3) In clustered index leaf page records, the number of fields must reside between the number of 'core' fields (dict_index_t::n_core_fields introduced in this work) and dict_index_t::n_fields. If the number of fields is less than dict_index_t::n_fields, the missing fields are replaced with the column value of the 'default row'. Note: The number of fields in the record may shrink if some of the last instantly added columns are updated to the value that is in the 'default row'. The function btr_cur_trim() implements this 'compression' on update and rollback; dtuple::trim() implements it on insert. (4) In ROW_FORMAT=COMPACT and ROW_FORMAT=DYNAMIC records, the new status value REC_STATUS_COLUMNS_ADDED will indicate the presence of a new record header that will encode n_fields-n_core_fields-1 in 1 or 2 bytes. (In ROW_FORMAT=REDUNDANT records, the record header always explicitly encodes the number of fields.) We introduce the undo log record type TRX_UNDO_INSERT_DEFAULT for covering the insert of the 'default row' record when instant ADD COLUMN is used for the first time. Subsequent instant ADD COLUMN can use TRX_UNDO_UPD_EXIST_REC. This is joint work with Vin Chen (陈福荣) from Tencent. The design that was discussed in April 2017 would not have allowed import or export of data files, because instead of the 'default row' it would have introduced a data dictionary table. The test rpl.rpl_alter_instant is exactly as contributed in pull request #408. The test innodb.instant_alter is based on a contributed test. The redo log record format changes for ROW_FORMAT=DYNAMIC and ROW_FORMAT=COMPACT are as contributed. (With this change present, crash recovery from MariaDB 10.3.1 will fail in spectacular ways!) Also the semantics of higher-level redo log records that modify the PAGE_INSTANT field is changed. The redo log format version identifier was already changed to LOG_HEADER_FORMAT_CURRENT=103 in MariaDB 10.3.1. Everything else has been rewritten by me. Thanks to Elena Stepanova, the code has been tested extensively. When rolling back an instant ADD COLUMN operation, we must empty the PAGE_FREE list after deleting or shortening the 'default row' record, by calling either btr_page_empty() or btr_page_reorganize(). We must know the size of each entry in the PAGE_FREE list. If rollback left a freed copy of the 'default row' in the PAGE_FREE list, we would be unable to determine its size (if it is in ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC) because it would contain more fields than the rolled-back definition of the clustered index. UNIV_SQL_DEFAULT: A new special constant that designates an instantly added column that is not present in the clustered index record. len_is_stored(): Check if a length is an actual length. There are two magic length values: UNIV_SQL_DEFAULT, UNIV_SQL_NULL. dict_col_t::def_val: The 'default row' value of the column. If the column is not added instantly, def_val.len will be UNIV_SQL_DEFAULT. dict_col_t: Add the accessors is_virtual(), is_nullable(), is_instant(), instant_value(). dict_col_t::remove_instant(): Remove the 'instant ADD' status of a column. dict_col_t::name(const dict_table_t& table): Replaces dict_table_get_col_name(). dict_index_t::n_core_fields: The original number of fields. For secondary indexes and if instant ADD COLUMN has not been used, this will be equal to dict_index_t::n_fields. dict_index_t::n_core_null_bytes: Number of bytes needed to represent the null flags; usually equal to UT_BITS_IN_BYTES(n_nullable). dict_index_t::NO_CORE_NULL_BYTES: Magic value signalling that n_core_null_bytes was not initialized yet from the clustered index root page. dict_index_t: Add the accessors is_instant(), is_clust(), get_n_nullable(), instant_field_value(). dict_index_t::instant_add_field(): Adjust clustered index metadata for instant ADD COLUMN. dict_index_t::remove_instant(): Remove the 'instant ADD' status of a clustered index when the table becomes empty, or the very first instant ADD COLUMN operation is rolled back. dict_table_t: Add the accessors is_instant(), is_temporary(), supports_instant(). dict_table_t::instant_add_column(): Adjust metadata for instant ADD COLUMN. dict_table_t::rollback_instant(): Adjust metadata on the rollback of instant ADD COLUMN. prepare_inplace_alter_table_dict(): First create the ctx->new_table, and only then decide if the table really needs to be rebuilt. We must split the creation of table or index metadata from the creation of the dictionary table records and the creation of the data. In this way, we can transform a table-rebuilding operation into an instant ADD COLUMN operation. Dictionary objects will only be added to cache when table rebuilding or index creation is needed. The ctx->instant_table will never be added to cache. dict_table_t::add_to_cache(): Modified and renamed from dict_table_add_to_cache(). Do not modify the table metadata. Let the callers invoke dict_table_add_system_columns() and if needed, set can_be_evicted. dict_create_sys_tables_tuple(), dict_create_table_step(): Omit the system columns (which will now exist in the dict_table_t object already at this point). dict_create_table_step(): Expect the callers to invoke dict_table_add_system_columns(). pars_create_table(): Before creating the table creation execution graph, invoke dict_table_add_system_columns(). row_create_table_for_mysql(): Expect all callers to invoke dict_table_add_system_columns(). create_index_dict(): Replaces row_merge_create_index_graph(). innodb_update_n_cols(): Renamed from innobase_update_n_virtual(). Call my_error() if an error occurs. btr_cur_instant_init(), btr_cur_instant_init_low(), btr_cur_instant_root_init(): Load additional metadata from the clustered index and set dict_index_t::n_core_null_bytes. This is invoked when table metadata is first loaded into the data dictionary. dict_boot(): Initialize n_core_null_bytes for the four hard-coded dictionary tables. dict_create_index_step(): Initialize n_core_null_bytes. This is executed as part of CREATE TABLE. dict_index_build_internal_clust(): Initialize n_core_null_bytes to NO_CORE_NULL_BYTES if table->supports_instant(). row_create_index_for_mysql(): Initialize n_core_null_bytes for CREATE TEMPORARY TABLE. commit_cache_norebuild(): Call the code to rename or enlarge columns in the cache only if instant ADD COLUMN is not being used. (Instant ADD COLUMN would copy all column metadata from instant_table to old_table, including the names and lengths.) PAGE_INSTANT: A new 13-bit field for storing dict_index_t::n_core_fields. This is repurposing the 16-bit field PAGE_DIRECTION, of which only the least significant 3 bits were used. The original byte containing PAGE_DIRECTION will be accessible via the new constant PAGE_DIRECTION_B. page_get_instant(), page_set_instant(): Accessors for the PAGE_INSTANT. page_ptr_get_direction(), page_get_direction(), page_ptr_set_direction(): Accessors for PAGE_DIRECTION. page_direction_reset(): Reset PAGE_DIRECTION, PAGE_N_DIRECTION. page_direction_increment(): Increment PAGE_N_DIRECTION and set PAGE_DIRECTION. rec_get_offsets(): Use the 'leaf' parameter for non-debug purposes, and assume that heap_no is always set. Initialize all dict_index_t::n_fields for ROW_FORMAT=REDUNDANT records, even if the record contains fewer fields. rec_offs_make_valid(): Add the parameter 'leaf'. rec_copy_prefix_to_dtuple(): Assert that the tuple is only built on the core fields. Instant ADD COLUMN only applies to the clustered index, and we should never build a search key that has more than the PRIMARY KEY and possibly DB_TRX_ID,DB_ROLL_PTR. All these columns are always present. dict_index_build_data_tuple(): Remove assertions that would be duplicated in rec_copy_prefix_to_dtuple(). rec_init_offsets(): Support ROW_FORMAT=REDUNDANT records whose number of fields is between n_core_fields and n_fields. cmp_rec_rec_with_match(): Implement the comparison between two MIN_REC_FLAG records. trx_t::in_rollback: Make the field available in non-debug builds. trx_start_for_ddl_low(): Remove dangerous error-tolerance. A dictionary transaction must be flagged as such before it has generated any undo log records. This is because trx_undo_assign_undo() will mark the transaction as a dictionary transaction in the undo log header right before the very first undo log record is being written. btr_index_rec_validate(): Account for instant ADD COLUMN row_undo_ins_remove_clust_rec(): On the rollback of an insert into SYS_COLUMNS, revert instant ADD COLUMN in the cache by removing the last column from the table and the clustered index. row_search_on_row_ref(), row_undo_mod_parse_undo_rec(), row_undo_mod(), trx_undo_update_rec_get_update(): Handle the 'default row' as a special case. dtuple_t::trim(index): Omit a redundant suffix of an index tuple right before insert or update. After instant ADD COLUMN, if the last fields of a clustered index tuple match the 'default row', there is no need to store them. While trimming the entry, we must hold a page latch, so that the table cannot be emptied and the 'default row' be deleted. btr_cur_optimistic_update(), btr_cur_pessimistic_update(), row_upd_clust_rec_by_insert(), row_ins_clust_index_entry_low(): Invoke dtuple_t::trim() if needed. row_ins_clust_index_entry(): Restore dtuple_t::n_fields after calling row_ins_clust_index_entry_low(). rec_get_converted_size(), rec_get_converted_size_comp(): Allow the number of fields to be between n_core_fields and n_fields. Do not support infimum,supremum. They are never supposed to be stored in dtuple_t, because page creation nowadays uses a lower-level method for initializing them. rec_convert_dtuple_to_rec_comp(): Assign the status bits based on the number of fields. btr_cur_trim(): In an update, trim the index entry as needed. For the 'default row', handle rollback specially. For user records, omit fields that match the 'default row'. btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete(): Skip locking and adaptive hash index for the 'default row'. row_log_table_apply_convert_mrec(): Replace 'default row' values if needed. In the temporary file that is applied by row_log_table_apply(), we must identify whether the records contain the extra header for instantly added columns. For now, we will allocate an additional byte for this for ROW_T_INSERT and ROW_T_UPDATE records when the source table has been subject to instant ADD COLUMN. The ROW_T_DELETE records are fine, as they will be converted and will only contain 'core' columns (PRIMARY KEY and some system columns) that are converted from dtuple_t. rec_get_converted_size_temp(), rec_init_offsets_temp(), rec_convert_dtuple_to_temp(): Add the parameter 'status'. REC_INFO_DEFAULT_ROW = REC_INFO_MIN_REC_FLAG | REC_STATUS_COLUMNS_ADDED: An info_bits constant for distinguishing the 'default row' record. rec_comp_status_t: An enum of the status bit values. rec_leaf_format: An enum that replaces the bool parameter of rec_init_offsets_comp_ordinary().
8 years ago
MDEV-16678 Prefer MDL to dict_sys.latch for innodb background tasks This is joint work with Thirunarayanan Balathandayuthapani. The MDL interface between InnoDB and the rest of the server (in storage/innobase/dict/dict0dict.cc and in include/) is my work, while most everything else is Thiru's. The collection of InnoDB persistent statistics and the defragmentation were not refactored to use MDL. They will keep relying on lower-level interlocking with fil_check_pending_operations(). The purge of transaction history and the background operations on fulltext indexes will use MDL. We will revert commit 2c4844c9e76427525e8c39a2d72686085efe89c3 (MDEV-17813) because thanks to MDL, purge cannot conflict with DDL operations anymore. For a similar reason, we will remove the MDEV-16222 test case from gcol.innodb_virtual_debug_purge. Purge is essentially replacing all use of the global dict_sys.latch with MDL. Purge will skip the undo log records for tables whose names start with #sql-ib or #sql2. Theoretically, such tables might be renamed back to visible table names if TRUNCATE fails to create a new table, or the final rename in ALTER TABLE...ALGORITHM=COPY fails. In that case, purge could permanently leave some garbage in the table. Such garbage will be tolerated; the table would not be considered corrupted. To avoid repeated MDL releases and acquisitions, trx_purge_attach_undo_recs() will sort undo log records by table_id, and purge_node_t will keep the MDL and table handle open for multiple successive undo log records. get_purge_table(): A new accessor, used during the purge of history for indexed virtual columns. This interface should ideally not exist at all. thd_mdl_context(): Accessor of THD::mdl_context. Wrapped in a new thd_mdl_service. dict_get_db_name_len(): Define inline. dict_acquire_mdl_shared(): Acquire explicit shared MDL on a table name if needed. dict_table_open_on_id(): Return MDL_ticket, if requested. dict_table_close(): Release MDL ticket, if requested. dict_fts_index_syncing(), dict_index_t::index_fts_syncing: Remove. row_drop_table_for_mysql() no longer needs to check these, because MDL guarantees that a fulltext index sync will not be in progress while MDL_EXCLUSIVE is protecting a DDL operation. dict_table_t::parse_name(): Parse the table name for acquiring MDL. purge_node_t::undo_recs: Change the type to std::list<trx_purge_rec_t*> (different container, and storing also roll_ptr). purge_node_t: Add mdl_ticket, last_table_id, purge_thd, mdl_hold_recs for acquiring MDL and for keeping the table open across multiple undo log records. purge_vcol_info_t, row_purge_store_vsec_cur(), row_purge_restore_vsec_cur(): Remove. We will acquire the MDL earlier. purge_sys_t::heap: Added, for reading undo log records. fts_sync_during_ddl(): Invoked during ALGORITHM=INPLACE operations to ensure that fts_sync_table() will not conflict with MDL_EXCLUSIVE. Uses fts_t::sync_message for bookkeeping.
6 years ago
MDEV-12266: Change dict_table_t::space to fil_space_t* InnoDB always keeps all tablespaces in the fil_system cache. The fil_system.LRU is only for closing file handles; the fil_space_t and fil_node_t for all data files will remain in main memory. Between startup to shutdown, they can only be created and removed by DDL statements. Therefore, we can let dict_table_t::space point directly to the fil_space_t. dict_table_t::space_id: A numeric tablespace ID for the corner cases where we do not have a tablespace. The most prominent examples are ALTER TABLE...DISCARD TABLESPACE or a missing or corrupted file. There are a few functional differences; most notably: (1) DROP TABLE will delete matching .ibd and .cfg files, even if they were not attached to the data dictionary. (2) Some error messages will report file names instead of numeric IDs. There still are many functions that use numeric tablespace IDs instead of fil_space_t*, and many functions could be converted to fil_space_t member functions. Also, Tablespace and Datafile should be merged with fil_space_t and fil_node_t. page_id_t and buf_page_get_gen() could use fil_space_t& instead of a numeric ID, and after moving to a single buffer pool (MDEV-15058), buf_pool_t::page_hash could be moved to fil_space_t::page_hash. FilSpace: Remove. Only few calls to fil_space_acquire() will remain, and gradually they should be removed. mtr_t::set_named_space_id(ulint): Renamed from set_named_space(), to prevent accidental calls to this slower function. Very few callers remain. fseg_create(), fsp_reserve_free_extents(): Take fil_space_t* as a parameter instead of a space_id. fil_space_t::rename(): Wrapper for fil_rename_tablespace_check(), fil_name_write_rename(), fil_rename_tablespace(). Mariabackup passes the parameter log=false; InnoDB passes log=true. dict_mem_table_create(): Take fil_space_t* instead of space_id as parameter. dict_process_sys_tables_rec_and_mtr_commit(): Replace the parameter 'status' with 'bool cached'. dict_get_and_save_data_dir_path(): Avoid copying the fil_node_t::name. fil_ibd_open(): Return the tablespace. fil_space_t::set_imported(): Replaces fil_space_set_imported(). truncate_t: Change many member function parameters to fil_space_t*, and remove page_size parameters. row_truncate_prepare(): Merge to its only caller. row_drop_table_from_cache(): Assert that the table is persistent. dict_create_sys_indexes_tuple(): Write SYS_INDEXES.SPACE=FIL_NULL if the tablespace has been discarded. row_import_update_discarded_flag(): Remove a constant parameter.
8 years ago
MDEV-11415 Remove excessive undo logging during ALTER TABLE…ALGORITHM=COPY If a crash occurs during ALTER TABLE…ALGORITHM=COPY, InnoDB would spend a lot of time rolling back writes to the intermediate copy of the table. To reduce the amount of busy work done, a work-around was introduced in commit fd069e2bb36a3c1c1f26d65dd298b07e6d83ac8b in MySQL 4.1.8 and 5.0.2, to commit the transaction after every 10,000 inserted rows. A proper fix would have been to disable the undo logging altogether and to simply drop the intermediate copy of the table on subsequent server startup. This is what happens in MariaDB 10.3 with MDEV-14717,MDEV-14585. In MariaDB 10.2, the intermediate copy of the table would be left behind with a name starting with the string #sql. This is a backport of a bug fix from MySQL 8.0.0 to MariaDB, contributed by jixianliang <271365745@qq.com>. Unlike recent MySQL, MariaDB supports ALTER IGNORE. For that operation InnoDB must for now keep the undo logging enabled, so that the latest row can be rolled back in case of an error. In Galera cluster, the LOAD DATA statement will retain the existing behaviour and commit the transaction after every 10,000 rows if the parameter wsrep_load_data_splitting=ON is set. The logic to do so (the wsrep_load_data_split() function and the call handler::extra(HA_EXTRA_FAKE_START_STMT)) are joint work by Ji Xianliang and Marko Mäkelä. The original fix: Author: Thirunarayanan Balathandayuthapani <thirunarayanan.balathandayuth@oracle.com> Date: Wed Dec 2 16:09:15 2015 +0530 Bug#17479594 AVOID INTERMEDIATE COMMIT WHILE DOING ALTER TABLE ALGORITHM=COPY Problem: During ALTER TABLE, we commit and restart the transaction for every 10,000 rows, so that the rollback after recovery would not take so long. Fix: Suppress the undo logging during copy alter operation. If fts_index is present then insert directly into fts auxiliary table rather than doing at commit time. ha_innobase::num_write_row: Remove the variable. ha_innobase::write_row(): Remove the hack for committing every 10000 rows. row_lock_table_for_mysql(): Remove the extra 2 parameters. lock_get_src_table(), lock_is_table_exclusive(): Remove. Reviewed-by: Marko Mäkelä <marko.makela@oracle.com> Reviewed-by: Shaohua Wang <shaohua.wang@oracle.com> Reviewed-by: Jon Olav Hauglid <jon.hauglid@oracle.com>
8 years ago
Merge 10.1 into 10.2 This only merges MDEV-12253, adapting it to MDEV-12602 which is already present in 10.2 but not yet in the 10.1 revision that is being merged. TODO: Error handling in crash recovery needs to be improved. If a page cannot be decrypted (or read), we should cleanly abort the startup. If innodb_force_recovery is specified, we should ignore the problematic page and apply redo log to other pages. Currently, the test encryption.innodb-redo-badkey randomly fails like this (the last messages are from cmake -DWITH_ASAN): 2017-05-05 10:19:40 140037071685504 [Note] InnoDB: Starting crash recovery from checkpoint LSN=1635994 2017-05-05 10:19:40 140037071685504 [ERROR] InnoDB: Missing MLOG_FILE_NAME or MLOG_FILE_DELETE before MLOG_CHECKPOINT for tablespace 1 2017-05-05 10:19:40 140037071685504 [ERROR] InnoDB: Plugin initialization aborted at srv0start.cc[2201] with error Data structure corruption 2017-05-05 10:19:41 140037071685504 [Note] InnoDB: Starting shutdown... i================================================================= ==5226==ERROR: AddressSanitizer: attempting free on address which was not malloc()-ed: 0x612000018588 in thread T0 #0 0x736750 in operator delete(void*) (/mariadb/server/build/sql/mysqld+0x736750) #1 0x1e4833f in LatchCounter::~LatchCounter() /mariadb/server/storage/innobase/include/sync0types.h:599:4 #2 0x1e480b8 in LatchMeta<LatchCounter>::~LatchMeta() /mariadb/server/storage/innobase/include/sync0types.h:786:17 #3 0x1e35509 in sync_latch_meta_destroy() /mariadb/server/storage/innobase/sync/sync0debug.cc:1622:3 #4 0x1e35314 in sync_check_close() /mariadb/server/storage/innobase/sync/sync0debug.cc:1839:2 #5 0x1dfdc18 in innodb_shutdown() /mariadb/server/storage/innobase/srv/srv0start.cc:2888:2 #6 0x197e5e6 in innobase_init(void*) /mariadb/server/storage/innobase/handler/ha_innodb.cc:4475:3
9 years ago
MDEV-6076 Persistent AUTO_INCREMENT for InnoDB This should be functionally equivalent to WL#6204 in MySQL 8.0.0, with the notable difference that the file format changes are limited to repurposing a previously unused data field in B-tree pages. For persistent InnoDB tables, write the last used AUTO_INCREMENT value to the root page of the clustered index, in the previously unused (0) PAGE_MAX_TRX_ID field, now aliased as PAGE_ROOT_AUTO_INC. Unlike some other previously unused InnoDB data fields, this one was actually always zero-initialized, at least since MySQL 3.23.49. The writes to PAGE_ROOT_AUTO_INC are protected by SX or X latch on the root page. The SX latch will allow concurrent read access to the root page. (The field PAGE_ROOT_AUTO_INC will only be read on the first-time call to ha_innobase::open() from the SQL layer. The PAGE_ROOT_AUTO_INC can only be updated when executing SQL, so read/write races are not possible.) During INSERT, the PAGE_ROOT_AUTO_INC is updated by the low-level function btr_cur_search_to_nth_level(), adding no extra page access. [Adaptive hash index lookup will be disabled during INSERT.] If some rare UPDATE modifies an AUTO_INCREMENT column, the PAGE_ROOT_AUTO_INC will be adjusted in a separate mini-transaction in ha_innobase::update_row(). When a page is reorganized, we have to preserve the PAGE_ROOT_AUTO_INC field. During ALTER TABLE, the initial AUTO_INCREMENT value will be copied from the table. ALGORITHM=COPY and online log apply in LOCK=NONE will update PAGE_ROOT_AUTO_INC in real time. innodb_col_no(): Determine the dict_table_t::cols[] element index corresponding to a Field of a non-virtual column. (The MySQL 5.7 implementation of virtual columns breaks the 1:1 relationship between Field::field_index and dict_table_t::cols[]. Virtual columns are omitted from dict_table_t::cols[]. Therefore, we must translate the field_index of AUTO_INCREMENT columns into an index of dict_table_t::cols[].) Upgrade from old data files: By default, the AUTO_INCREMENT sequence in old data files would appear to be reset, because PAGE_MAX_TRX_ID or PAGE_ROOT_AUTO_INC would contain the value 0 in each clustered index page. In new data files, PAGE_ROOT_AUTO_INC can only be 0 if the table is empty or does not contain any AUTO_INCREMENT column. For backward compatibility, we use the old method of SELECT MAX(auto_increment_column) for initializing the sequence. btr_read_autoinc(): Read the AUTO_INCREMENT sequence from a new-format data file. btr_read_autoinc_with_fallback(): A variant of btr_read_autoinc() that will resort to reading MAX(auto_increment_column) for data files that did not use AUTO_INCREMENT yet. It was manually tested that during the execution of innodb.autoinc_persist the compatibility logic is not activated (for new files, PAGE_ROOT_AUTO_INC is never 0 in nonempty clustered index root pages). initialize_auto_increment(): Replaces ha_innobase::innobase_initialize_autoinc(). This initializes the AUTO_INCREMENT metadata. Only called from ha_innobase::open(). ha_innobase::info_low(): Do not try to lazily initialize dict_table_t::autoinc. It must already have been initialized by ha_innobase::open() or ha_innobase::create(). Note: The adjustments to class ha_innopart were not tested, because the source code (native InnoDB partitioning) is not being compiled.
9 years ago
MDEV-15662 Instant DROP COLUMN or changing the order of columns Allow ADD COLUMN anywhere in a table, not only adding as the last column. Allow instant DROP COLUMN and instant changing the order of columns. The added columns will always be added last in clustered index records. In new records, instantly dropped columns will be stored as NULL or empty when possible. Information about dropped and reordered columns will be written in a metadata BLOB (mblob), which is stored before the first 'user' field in the hidden metadata record at the start of the clustered index. The presence of mblob is indicated by setting the delete-mark flag in the metadata record. The metadata BLOB stores the number of clustered index fields, followed by an array of column information for each field. For dropped columns, we store the NOT NULL flag, the fixed length, and for variable-length columns, whether the maximum length exceeded 255 bytes. For non-dropped columns, we store the column position. Unlike with MDEV-11369, when a table becomes empty, it cannot be converted back to the canonical format. The reason for this is that other threads may hold cached objects such as row_prebuilt_t::ins_node that could refer to dropped or reordered index fields. For instant DROP COLUMN and ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC, we must store the n_core_null_bytes in the root page, so that the chain of node pointer records can be followed in order to reach the leftmost leaf page where the metadata record is located. If the mblob is present, we will zero-initialize the strings "infimum" and "supremum" in the root page, and use the last byte of "supremum" for storing the number of null bytes (which are allocated but useless on node pointer pages). This is necessary for btr_cur_instant_init_metadata() to be able to navigate to the mblob. If the PRIMARY KEY contains any variable-length column and some nullable columns were instantly dropped, the dict_index_t::n_nullable in the data dictionary could be smaller than it actually is in the non-leaf pages. Because of this, the non-leaf pages could use more bytes for the null flags than the data dictionary expects, and we could be reading the lengths of the variable-length columns from the wrong offset, and thus reading the child page number from wrong place. This is the result of two design mistakes that involve unnecessary storage of data: First, it is nonsense to store any data fields for the leftmost node pointer records, because the comparisons would be resolved by the MIN_REC_FLAG alone. Second, there cannot be any null fields in the clustered index node pointer fields, but we nevertheless reserve space for all the null flags. Limitations (future work): MDEV-17459 Allow instant ALTER TABLE even if FULLTEXT INDEX exists MDEV-17468 Avoid table rebuild on operations on generated columns MDEV-17494 Refuse ALGORITHM=INSTANT when the row size is too large btr_page_reorganize_low(): Preserve any metadata in the root page. Call lock_move_reorganize_page() only after restoring the "infimum" and "supremum" records, to avoid a memcmp() assertion failure. dict_col_t::DROPPED: Magic value for dict_col_t::ind. dict_col_t::clear_instant(): Renamed from dict_col_t::remove_instant(). Do not assert that the column was instantly added, because we sometimes call this unconditionally for all columns. Convert an instantly added column to a "core column". The old name remove_instant() could be mistaken to refer to "instant DROP COLUMN". dict_col_t::is_added(): Rename from dict_col_t::is_instant(). dtype_t::metadata_blob_init(): Initialize the mblob data type. dtuple_t::is_metadata(), dtuple_t::is_alter_metadata(), upd_t::is_metadata(), upd_t::is_alter_metadata(): Check if info_bits refer to a metadata record. dict_table_t::instant: Metadata about dropped or reordered columns. dict_table_t::prepare_instant(): Prepare ha_innobase_inplace_ctx::instant_table for instant ALTER TABLE. innobase_instant_try() will pass this to dict_table_t::instant_column(). On rollback, dict_table_t::rollback_instant() will be called. dict_table_t::instant_column(): Renamed from instant_add_column(). Add the parameter col_map so that columns can be reordered. Copy and adjust v_cols[] as well. dict_table_t::find(): Find an old column based on a new column number. dict_table_t::serialise_columns(), dict_table_t::deserialise_columns(): Convert the mblob. dict_index_t::instant_metadata(): Create the metadata record for instant ALTER TABLE. Invoke dict_table_t::serialise_columns(). dict_index_t::reconstruct_fields(): Invoked by dict_table_t::deserialise_columns(). dict_index_t::clear_instant_alter(): Move the fields for the dropped columns to the end, and sort the surviving index fields in ascending order of column position. ha_innobase::check_if_supported_inplace_alter(): Do not allow adding a FTS_DOC_ID column if a hidden FTS_DOC_ID column exists due to FULLTEXT INDEX. (This always required ALGORITHM=COPY.) instant_alter_column_possible(): Add a parameter for InnoDB table, to check for additional conditions, such as the maximum number of index fields. ha_innobase_inplace_ctx::first_alter_pos: The first column whose position is affected by instant ADD, DROP, or changing the order of columns. innobase_build_col_map(): Skip added virtual columns. prepare_inplace_add_virtual(): Correctly compute num_to_add_vcol. Remove some unnecessary code. Note that the call to innodb_base_col_setup() should be executed later. commit_try_norebuild(): If ctx->is_instant(), let the virtual columns be added or dropped by innobase_instant_try(). innobase_instant_try(): Fill in a zero default value for the hidden column FTS_DOC_ID (to reduce the work needed in MDEV-17459). If any columns were dropped or reordered (or added not last), delete any SYS_COLUMNS records for the following columns, and insert SYS_COLUMNS records for all subsequent stored columns as well as for all virtual columns. If any virtual column is dropped, rewrite all virtual column metadata. Use a shortcut only for adding virtual columns. This is because innobase_drop_virtual_try() assumes that the dropped virtual columns still exist in ctx->old_table. innodb_update_cols(): Renamed from innodb_update_n_cols(). innobase_add_one_virtual(), innobase_insert_sys_virtual(): Change the return type to bool, and invoke my_error() when detecting an error. innodb_insert_sys_columns(): Insert a record into SYS_COLUMNS. Refactored from innobase_add_one_virtual() and innobase_instant_add_col(). innobase_instant_add_col(): Replace the parameter dfield with type. innobase_instant_drop_cols(): Drop matching columns from SYS_COLUMNS and all columns from SYS_VIRTUAL. innobase_add_virtual_try(), innobase_drop_virtual_try(): Let the caller invoke innodb_update_cols(). innobase_rename_column_try(): Skip dropped columns. commit_cache_norebuild(): Update table->fts->doc_col. dict_mem_table_col_rename_low(): Skip dropped columns. trx_undo_rec_get_partial_row(): Skip dropped columns. trx_undo_update_rec_get_update(): Handle the metadata BLOB correctly. trx_undo_page_report_modify(): Avoid out-of-bounds access to record fields. Log metadata records consistently. Apparently, the first fields of a clustered index may be updated in an update_undo vector when the index is ID_IND of SYS_FOREIGN, as part of renaming the table during ALTER TABLE. Normally, updates of the PRIMARY KEY should be logged as delete-mark and an insert. row_undo_mod_parse_undo_rec(), row_purge_parse_undo_rec(): Use trx_undo_metadata. row_undo_mod_clust_low(): On metadata rollback, roll back the root page too. row_undo_mod_clust(): Relax an assertion. The delete-mark flag was repurposed for ALTER TABLE metadata records. row_rec_to_index_entry_impl(): Add the template parameter mblob and the optional parameter info_bits for specifying the desired new info bits. For the metadata tuple, allow conversion between the original format (ADD COLUMN only) and the generic format (with hidden BLOB). Add the optional parameter "pad" to determine whether the tuple should be padded to the index fields (on ALTER TABLE it should), or whether it should remain at its original size (on rollback). row_build_index_entry_low(): Clean up the code, removing redundant variables and conditions. For instantly dropped columns, generate a dummy value that is NULL, the empty string, or a fixed length of NUL bytes, depending on the type of the dropped column. row_upd_clust_rec_by_insert_inherit_func(): On the update of PRIMARY KEY of a record that contained a dropped column whose value was stored externally, we will be inserting a dummy NULL or empty string value to the field of the dropped column. The externally stored column would eventually be dropped when purge removes the delete-marked record for the old PRIMARY KEY value. btr_index_rec_validate(): Recognize the metadata record. btr_discard_only_page_on_level(): Preserve the generic instant ALTER TABLE metadata. btr_set_instant(): Replaces page_set_instant(). This sets a clustered index root page to the appropriate format, or upgrades from the MDEV-11369 instant ADD COLUMN to generic ALTER TABLE format. btr_cur_instant_init_low(): Read and validate the metadata BLOB page before reconstructing the dictionary information based on it. btr_cur_instant_init_metadata(): Do not read any lengths from the metadata record header before reading the BLOB. At this point, we would not actually know how many nullable fields the metadata record contains. btr_cur_instant_root_init(): Initialize n_core_null_bytes in one of two possible ways. btr_cur_trim(): Handle the mblob record. row_metadata_to_tuple(): Convert a metadata record to a data tuple, based on the new info_bits of the metadata record. btr_cur_pessimistic_update(): Invoke row_metadata_to_tuple() if needed. Invoke dtuple_convert_big_rec() for metadata records if the record is too large, or if the mblob is not yet marked as externally stored. btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete(): When the last user record is deleted, do not delete the generic instant ALTER TABLE metadata record. Only delete MDEV-11369 instant ADD COLUMN metadata records. btr_cur_optimistic_insert(): Avoid unnecessary computation of rec_size. btr_pcur_store_position(): Allow a logically empty page to contain a metadata record for generic ALTER TABLE. REC_INFO_DEFAULT_ROW_ADD: Renamed from REC_INFO_DEFAULT_ROW. This is for the old instant ADD COLUMN (MDEV-11369) only. REC_INFO_DEFAULT_ROW_ALTER: The more generic metadata record, with additional information for dropped or reordered columns. rec_info_bits_valid(): Remove. The only case when this would fail is when the record is the generic ALTER TABLE metadata record. rec_is_alter_metadata(): Check if a record is the metadata record for instant ALTER TABLE (other than ADD COLUMN). NOTE: This function must not be invoked on node pointer records, because the delete-mark flag in those records may be set (it is garbage), and then a debug assertion could fail because index->is_instant() does not necessarily hold. rec_is_add_metadata(): Check if a record is MDEV-11369 ADD COLUMN metadata record (not more generic instant ALTER TABLE). rec_get_converted_size_comp_prefix_low(): Assume that the metadata field will be stored externally. In dtuple_convert_big_rec() during the rec_get_converted_size() call, it would not be there yet. rec_get_converted_size_comp(): Replace status,fields,n_fields with tuple. rec_init_offsets_comp_ordinary(), rec_get_converted_size_comp_prefix_low(), rec_convert_dtuple_to_rec_comp(): Add template<bool mblob = false>. With mblob=true, process a record with a metadata BLOB. rec_copy_prefix_to_buf(): Assert that no fields beyond the key and system columns are being copied. Exclude the metadata BLOB field. rec_convert_dtuple_to_metadata_comp(): Convert an alter metadata tuple into a record. row_upd_index_replace_metadata(): Apply an update vector to an alter_metadata tuple. row_log_allocate(): Replace dict_index_t::is_instant() with a more appropriate condition that ignores dict_table_t::instant. Only a table on which the MDEV-11369 ADD COLUMN was performed can "lose its instantness" when it becomes empty. After instant DROP COLUMN or reordering columns, we cannot simply convert the table to the canonical format, because the data dictionary cache and all possibly existing references to it from other client connection threads would have to be adjusted. row_quiesce_write_index_fields(): Do not crash when the table contains an instantly dropped column. Thanks to Thirunarayanan Balathandayuthapani for discussing the design and implementing an initial prototype of this. Thanks to Matthias Leich for testing.
7 years ago
MDEV-22456 Dropping the adaptive hash index may cause DDL to lock up InnoDB If the InnoDB buffer pool contains many pages for a table or index that is being dropped or rebuilt, and if many of such pages are pointed to by the adaptive hash index, dropping the adaptive hash index may consume a lot of time. The time-consuming operation of dropping the adaptive hash index entries is being executed while the InnoDB data dictionary cache dict_sys is exclusively locked. It is not actually necessary to drop all adaptive hash index entries at the time a table or index is being dropped or rebuilt. We can let the LRU replacement policy of the buffer pool take care of this gradually. For this to work, we must detach the dict_table_t and dict_index_t objects from the main dict_sys cache, and once the last adaptive hash index entry for the detached table is removed (when the garbage page is evicted from the buffer pool) we can free the dict_table_t and dict_index_t object. Related to this, in MDEV-16283, we made ALTER TABLE...DISCARD TABLESPACE skip both the buffer pool eviction and the drop of the adaptive hash index. We shifted the burden to ALTER TABLE...IMPORT TABLESPACE or DROP TABLE. We can remove the eviction from DROP TABLE. We must retain the eviction in the ALTER TABLE...IMPORT TABLESPACE code path, so that in case the discarded table is being re-imported with the same tablespace identifier, the fresh data from the imported tablespace will replace any stale pages in the buffer pool. rpl.rpl_failed_drop_tbl_binlog: Remove the test. DROP TABLE can no longer be interrupted inside InnoDB. fseg_free_page(), fseg_free_step(), fseg_free_step_not_header(), fseg_free_page_low(), fseg_free_extent(): Remove the parameter that specifies whether the adaptive hash index should be dropped. btr_search_lazy_free(): Lazily free an index when the last reference to it is dropped from the adaptive hash index. buf_pool_clear_hash_index(): Declare static, and move to the same compilation unit with the bulk of the adaptive hash index code. dict_index_t::clone(), dict_index_t::clone_if_needed(): Clone an index that is being rebuilt while adaptive hash index entries exist. The original index will be inserted into dict_table_t::freed_indexes and dict_index_t::set_freed() will be called. dict_index_t::set_freed(), dict_index_t::freed(): Note that or check whether the index has been freed. We will use the impossible page number 1 to denote this condition. dict_index_t::n_ahi_pages(): Replaces btr_search_info_get_ref_count(). dict_index_t::detach_columns(): Move the assignment n_fields=0 to ha_innobase_inplace_ctx::clear_added_indexes(). We must have access to the columns when freeing the adaptive hash index. Note: dict_table_t::v_cols[] will remain valid. If virtual columns are dropped or added, the table definition will be reloaded in ha_innobase::commit_inplace_alter_table(). buf_page_mtr_lock(): Drop a stale adaptive hash index if needed. We will also reduce the number of btr_get_search_latch() calls and enclose some more code inside #ifdef BTR_CUR_HASH_ADAPT in order to benefit cmake -DWITH_INNODB_AHI=OFF.
5 years ago
MDEV-22456 Dropping the adaptive hash index may cause DDL to lock up InnoDB If the InnoDB buffer pool contains many pages for a table or index that is being dropped or rebuilt, and if many of such pages are pointed to by the adaptive hash index, dropping the adaptive hash index may consume a lot of time. The time-consuming operation of dropping the adaptive hash index entries is being executed while the InnoDB data dictionary cache dict_sys is exclusively locked. It is not actually necessary to drop all adaptive hash index entries at the time a table or index is being dropped or rebuilt. We can let the LRU replacement policy of the buffer pool take care of this gradually. For this to work, we must detach the dict_table_t and dict_index_t objects from the main dict_sys cache, and once the last adaptive hash index entry for the detached table is removed (when the garbage page is evicted from the buffer pool) we can free the dict_table_t and dict_index_t object. Related to this, in MDEV-16283, we made ALTER TABLE...DISCARD TABLESPACE skip both the buffer pool eviction and the drop of the adaptive hash index. We shifted the burden to ALTER TABLE...IMPORT TABLESPACE or DROP TABLE. We can remove the eviction from DROP TABLE. We must retain the eviction in the ALTER TABLE...IMPORT TABLESPACE code path, so that in case the discarded table is being re-imported with the same tablespace identifier, the fresh data from the imported tablespace will replace any stale pages in the buffer pool. rpl.rpl_failed_drop_tbl_binlog: Remove the test. DROP TABLE can no longer be interrupted inside InnoDB. fseg_free_page(), fseg_free_step(), fseg_free_step_not_header(), fseg_free_page_low(), fseg_free_extent(): Remove the parameter that specifies whether the adaptive hash index should be dropped. btr_search_lazy_free(): Lazily free an index when the last reference to it is dropped from the adaptive hash index. buf_pool_clear_hash_index(): Declare static, and move to the same compilation unit with the bulk of the adaptive hash index code. dict_index_t::clone(), dict_index_t::clone_if_needed(): Clone an index that is being rebuilt while adaptive hash index entries exist. The original index will be inserted into dict_table_t::freed_indexes and dict_index_t::set_freed() will be called. dict_index_t::set_freed(), dict_index_t::freed(): Note that or check whether the index has been freed. We will use the impossible page number 1 to denote this condition. dict_index_t::n_ahi_pages(): Replaces btr_search_info_get_ref_count(). dict_index_t::detach_columns(): Move the assignment n_fields=0 to ha_innobase_inplace_ctx::clear_added_indexes(). We must have access to the columns when freeing the adaptive hash index. Note: dict_table_t::v_cols[] will remain valid. If virtual columns are dropped or added, the table definition will be reloaded in ha_innobase::commit_inplace_alter_table(). buf_page_mtr_lock(): Drop a stale adaptive hash index if needed. We will also reduce the number of btr_get_search_latch() calls and enclose some more code inside #ifdef BTR_CUR_HASH_ADAPT in order to benefit cmake -DWITH_INNODB_AHI=OFF.
5 years ago
11 years ago
MDEV-24971 InnoDB access freed virtual column after rollback of secondary index Problem: ======== InnoDB fails to clean the index stub if it fails to add the virtual index which contains new virtual column. But it clears the newly virtual column from index in clear_added_indexes() during inplace_alter_table. On commit, InnoDB evicts and reload the table. In case of rollback, it doesn't happen. InnoDB clears the ABORTED index while opening the table or doing the DDL. In the mean time, InnoDB can access the dropped virtual index columns while creating prebuilt or rollback of concurrent DML. Solution: ========== (1) InnoDB should maintain newly added virtual column while rollbacking the newly added virtual index. (2) InnoDB must not defer the index removal if the alter table is executed with LOCK=EXCLUSIVE. (3) For LOCK=SHARED, InnoDB should check whether the table has any other transaction lock other than alter transaction before deferring the index stub. Replaced has_new_v_col with dict_add_vcol_info in dict_index_t to indicate whether the index has any new virtual column. dict_index_t::has_new_v_col(): Returns whether the index has newly added virtual column, it doesn't say which columns are newly added virtual column ha_innobase_inplace_ctx::is_new_vcol(): Return whether the given column is added as a part of the current alter. ha_innobase_inplace_ctx::clean_new_vcol_index(): Copy the newly added virtual column to new_vcol_info in dict_index_t. Replace the column in the index fields with virtual column stored in new_vcol_info. dict_index_t::assign_new_v_col(): Store the number of virtual column added in index as a part of alter table. dict_index_t::get_n_new_vcol(): Get the number of newly added virtual column dict_index_t::assign_drop_v_col(): Allocate the memory for adding new virtual column in new_vcol_info. dict_index_t::add_drop_v_col(): Add the newly added virtual column in new_vcol_info. dict_table_t::has_lock_for_other_trx(): Whether the table has any other transaction lock than given transaction. row_merge_drop_indexes(): Add parameter alter_trx and check whether the table has any other lock than alter transaction.
5 years ago
MDEV-25004 Missing row in FTS_DOC_ID_INDEX during DELETE HISTORY 1. In case of system-versioned table add row_end into FTS_DOC_ID index in fts_create_common_tables() and innobase_create_key_defs(). fts_n_uniq() returns 1 or 2 depending on whether the table is system-versioned. After this patch recreate of FTS_DOC_ID index is required for existing system-versioned tables. If you see this message in error log or server warnings: "InnoDB: Table db/t1 contains 2 indexes inside InnoDB, which is different from the number of indexes 1 defined in the MariaDB" use this command to fix the table: ALTER TABLE db.t1 FORCE; 2. Fix duplicate history for secondary unique index like it was done in MDEV-23644 for clustered index (932ec586aad). In case of existing history row which conflicts with currently inseted row we check in row_ins_scan_sec_index_for_duplicate() whether that row was inserted as part of current transaction. In that case we indicate with DB_FOREIGN_DUPLICATE_KEY that new history row is not needed and should be silently skipped. 3. Some parts of MDEV-21138 (7410ff436e9) reverted. Skipping of FTS_DOC_ID index for history rows made problems with purge system. Now this is fixed differently by p.2. 4. wait_all_purged.inc checks that we didn't affect non-history rows so they are deleted and purged correctly. Additional FTS fixes fts_init_get_doc_id(): exclude history rows from max_doc_id calculation. fts_init_get_doc_id() callback is used only for crash recovery. fts_add_doc_by_id(): set max value for row_end field. fts_read_stopword(): stopwords table can be system-versioned too. We now read stopwords only for current data. row_insert_for_mysql(): exclude history rows from doc_id validation. row_merge_read_clustered_index(): exclude history_rows from doc_id processing. fts_load_user_stopword(): for versioned table retrieve row_end field and skip history rows. For non-versioned table we retrieve 'value' field twice (just for uniformity). FTS tests for System Versioning now include maybe_versioning.inc which adds 3 combinations: 'vers' for debug build sets sysvers_force and sysvers_hide. sysvers_force makes every created table system-versioned, sysvers_hide hides WITH SYSTEM VERSIONING for SHOW CREATE. Note: basic.test, stopword.test and versioning.test do not require debug for 'vers' combination. This is controlled by $modify_create_table in maybe_versioning.inc and these tests run WITH SYSTEM VERSIONING explicitly which allows to test 'vers' combination on non-debug builds. 'vers_trx' like 'vers' sets sysvers_force_trx and sysvers_hide. That tests FTS with trx_id-based System Versioning. 'orig' works like before: no System Versioning is added, no debug is required. Upgrade/downgrade test for System Versioning is done by innodb_fts.versioning. It has 2 combinations: 'prepare' makes binaries in std_data (requires old server and OLD_BINDIR). It tests upgrade/downgrade against old server as well. 'upgrade' tests upgrade against binaries in std_data. Cleanups: Removed innodb-fts-stopword.test as it duplicates stopword.test
3 years ago
MDEV-33277 In-place upgrade causes invalid AUTO_INCREMENT values MDEV-33308 CHECK TABLE is modifying .frm file even if --read-only As noted in commit d0ef1aaf618452efea5a2e74f6d86ed0a586b63e, MySQL as well as older versions of MariaDB server would during ALTER TABLE ... IMPORT TABLESPACE write bogus values to the PAGE_MAX_TRX_ID field to pages of the clustered index, instead of letting that field remain 0. In commit 8777458a6eb73ac1d7d864ebac390ea7039e21c1 this field was repurposed for PAGE_ROOT_AUTO_INC in the clustered index root page. To avoid trouble when upgrading from MySQL or older versions of MariaDB, we will try to detect and correct bogus values of PAGE_ROOT_AUTO_INC when opening a table for the first time from the SQL layer. btr_read_autoinc_with_fallback(): Add the parameters to mysql_version,max to indicate the TABLE_SHARE::mysql_version of the .frm file and the maximum value allowed for the type of the AUTO_INCREMENT column. In case the table was originally created in MySQL or an older version of MariaDB, read also the maximum value of the AUTO_INCREMENT column from the table and reset the PAGE_ROOT_AUTO_INC if it is above the limit. dict_table_t::get_index(const dict_col_t &) const: Find an index that starts with the specified column. ha_innobase::check_for_upgrade(): Return HA_ADMIN_FAILED if InnoDB needs upgrading but is in read-only mode. In this way, the call to update_frm_version() will be skipped. row_import_autoinc(): Adjust the AUTO_INCREMENT column at the end of ALTER TABLE...IMPORT TABLESPACE. This refinement was suggested by Debarun Banerjee. The changes outside InnoDB were developed by Michael 'Monty' Widenius: Added print_check_msg() service for easy reporting of check/repair messages in ENGINE=Aria and ENGINE=InnoDB. Fixed that CHECK TABLE do not update the .frm file under --read-only. Added 'handler_flags' to HA_CHECK_OPT as a way for storage engines to store state from handler::check_for_upgrade(). Reviewed by: Debarun Banerjee
2 years ago
MDEV-12253: Buffer pool blocks are accessed after they have been freed Problem was that bpage was referenced after it was already freed from LRU. Fixed by adding a new variable encrypted that is passed down to buf_page_check_corrupt() and used in buf_page_get_gen() to stop processing page read. This patch should also address following test failures and bugs: MDEV-12419: IMPORT should not look up tablespace in PageConverter::validate(). This is now removed. MDEV-10099: encryption.innodb_onlinealter_encryption fails sporadically in buildbot MDEV-11420: encryption.innodb_encryption-page-compression failed in buildbot MDEV-11222: encryption.encrypt_and_grep failed in buildbot on P8 Removed dict_table_t::is_encrypted and dict_table_t::ibd_file_missing and replaced these with dict_table_t::file_unreadable. Table ibd file is missing if fil_get_space(space_id) returns NULL and encrypted if not. Removed dict_table_t::is_corrupted field. Ported FilSpace class from 10.2 and using that on buf_page_check_corrupt(), buf_page_decrypt_after_read(), buf_page_encrypt_before_write(), buf_dblwr_process(), buf_read_page(), dict_stats_save_defrag_stats(). Added test cases when enrypted page could be read while doing redo log crash recovery. Also added test case for row compressed blobs. btr_cur_open_at_index_side_func(), btr_cur_open_at_rnd_pos_func(): Avoid referencing block that is NULL. buf_page_get_zip(): Issue error if page read fails. buf_page_get_gen(): Use dberr_t for error detection and do not reference bpage after we hare freed it. buf_mark_space_corrupt(): remove bpage from LRU also when it is encrypted. buf_page_check_corrupt(): @return DB_SUCCESS if page has been read and is not corrupted, DB_PAGE_CORRUPTED if page based on checksum check is corrupted, DB_DECRYPTION_FAILED if page post encryption checksum matches but after decryption normal page checksum does not match. In read case only DB_SUCCESS is possible. buf_page_io_complete(): use dberr_t for error handling. buf_flush_write_block_low(), buf_read_ahead_random(), buf_read_page_async(), buf_read_ahead_linear(), buf_read_ibuf_merge_pages(), buf_read_recv_pages(), fil_aio_wait(): Issue error if page read fails. btr_pcur_move_to_next_page(): Do not reference page if it is NULL. Introduced dict_table_t::is_readable() and dict_index_t::is_readable() that will return true if tablespace exists and pages read from tablespace are not corrupted or page decryption failed. Removed buf_page_t::key_version. After page decryption the key version is not removed from page frame. For unencrypted pages, old key_version is removed at buf_page_encrypt_before_write() dict_stats_update_transient_for_index(), dict_stats_update_transient() Do not continue if table decryption failed or table is corrupted. dict0stats.cc: Introduced a dict_stats_report_error function to avoid code duplication. fil_parse_write_crypt_data(): Check that key read from redo log entry is found from encryption plugin and if it is not, refuse to start. PageConverter::validate(): Removed access to fil_space_t as tablespace is not available during import. Fixed error code on innodb.innodb test. Merged test cased innodb-bad-key-change5 and innodb-bad-key-shutdown to innodb-bad-key-change2. Removed innodb-bad-key-change5 test. Decreased unnecessary complexity on some long lasting tests. Removed fil_inc_pending_ops(), fil_decr_pending_ops(), fil_get_first_space(), fil_get_next_space(), fil_get_first_space_safe(), fil_get_next_space_safe() functions. fil_space_verify_crypt_checksum(): Fixed bug found using ASAN where FIL_PAGE_END_LSN_OLD_CHECKSUM field was incorrectly accessed from row compressed tables. Fixed out of page frame bug for row compressed tables in fil_space_verify_crypt_checksum() found using ASAN. Incorrect function was called for compressed table. Added new tests for discard, rename table and drop (we should allow them even when page decryption fails). Alter table rename is not allowed. Added test for restart with innodb-force-recovery=1 when page read on redo-recovery cant be decrypted. Added test for corrupted table where both page data and FIL_PAGE_FILE_FLUSH_LSN_OR_KEY_VERSION is corrupted. Adjusted the test case innodb_bug14147491 so that it does not anymore expect crash. Instead table is just mostly not usable. fil0fil.h: fil_space_acquire_low is not visible function and fil_space_acquire and fil_space_acquire_silent are inline functions. FilSpace class uses fil_space_acquire_low directly. recv_apply_hashed_log_recs() does not return anything.
9 years ago
MDEV-11369 Instant ADD COLUMN for InnoDB For InnoDB tables, adding, dropping and reordering columns has required a rebuild of the table and all its indexes. Since MySQL 5.6 (and MariaDB 10.0) this has been supported online (LOCK=NONE), allowing concurrent modification of the tables. This work revises the InnoDB ROW_FORMAT=REDUNDANT, ROW_FORMAT=COMPACT and ROW_FORMAT=DYNAMIC so that columns can be appended instantaneously, with only minor changes performed to the table structure. The counter innodb_instant_alter_column in INFORMATION_SCHEMA.GLOBAL_STATUS is incremented whenever a table rebuild operation is converted into an instant ADD COLUMN operation. ROW_FORMAT=COMPRESSED tables will not support instant ADD COLUMN. Some usability limitations will be addressed in subsequent work: MDEV-13134 Introduce ALTER TABLE attributes ALGORITHM=NOCOPY and ALGORITHM=INSTANT MDEV-14016 Allow instant ADD COLUMN, ADD INDEX, LOCK=NONE The format of the clustered index (PRIMARY KEY) is changed as follows: (1) The FIL_PAGE_TYPE of the root page will be FIL_PAGE_TYPE_INSTANT, and a new field PAGE_INSTANT will contain the original number of fields in the clustered index ('core' fields). If instant ADD COLUMN has not been used or the table becomes empty, or the very first instant ADD COLUMN operation is rolled back, the fields PAGE_INSTANT and FIL_PAGE_TYPE will be reset to 0 and FIL_PAGE_INDEX. (2) A special 'default row' record is inserted into the leftmost leaf, between the page infimum and the first user record. This record is distinguished by the REC_INFO_MIN_REC_FLAG, and it is otherwise in the same format as records that contain values for the instantly added columns. This 'default row' always has the same number of fields as the clustered index according to the table definition. The values of 'core' fields are to be ignored. For other fields, the 'default row' will contain the default values as they were during the ALTER TABLE statement. (If the column default values are changed later, those values will only be stored in the .frm file. The 'default row' will contain the original evaluated values, which must be the same for every row.) The 'default row' must be completely hidden from higher-level access routines. Assertions have been added to ensure that no 'default row' is ever present in the adaptive hash index or in locked records. The 'default row' is never delete-marked. (3) In clustered index leaf page records, the number of fields must reside between the number of 'core' fields (dict_index_t::n_core_fields introduced in this work) and dict_index_t::n_fields. If the number of fields is less than dict_index_t::n_fields, the missing fields are replaced with the column value of the 'default row'. Note: The number of fields in the record may shrink if some of the last instantly added columns are updated to the value that is in the 'default row'. The function btr_cur_trim() implements this 'compression' on update and rollback; dtuple::trim() implements it on insert. (4) In ROW_FORMAT=COMPACT and ROW_FORMAT=DYNAMIC records, the new status value REC_STATUS_COLUMNS_ADDED will indicate the presence of a new record header that will encode n_fields-n_core_fields-1 in 1 or 2 bytes. (In ROW_FORMAT=REDUNDANT records, the record header always explicitly encodes the number of fields.) We introduce the undo log record type TRX_UNDO_INSERT_DEFAULT for covering the insert of the 'default row' record when instant ADD COLUMN is used for the first time. Subsequent instant ADD COLUMN can use TRX_UNDO_UPD_EXIST_REC. This is joint work with Vin Chen (陈福荣) from Tencent. The design that was discussed in April 2017 would not have allowed import or export of data files, because instead of the 'default row' it would have introduced a data dictionary table. The test rpl.rpl_alter_instant is exactly as contributed in pull request #408. The test innodb.instant_alter is based on a contributed test. The redo log record format changes for ROW_FORMAT=DYNAMIC and ROW_FORMAT=COMPACT are as contributed. (With this change present, crash recovery from MariaDB 10.3.1 will fail in spectacular ways!) Also the semantics of higher-level redo log records that modify the PAGE_INSTANT field is changed. The redo log format version identifier was already changed to LOG_HEADER_FORMAT_CURRENT=103 in MariaDB 10.3.1. Everything else has been rewritten by me. Thanks to Elena Stepanova, the code has been tested extensively. When rolling back an instant ADD COLUMN operation, we must empty the PAGE_FREE list after deleting or shortening the 'default row' record, by calling either btr_page_empty() or btr_page_reorganize(). We must know the size of each entry in the PAGE_FREE list. If rollback left a freed copy of the 'default row' in the PAGE_FREE list, we would be unable to determine its size (if it is in ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC) because it would contain more fields than the rolled-back definition of the clustered index. UNIV_SQL_DEFAULT: A new special constant that designates an instantly added column that is not present in the clustered index record. len_is_stored(): Check if a length is an actual length. There are two magic length values: UNIV_SQL_DEFAULT, UNIV_SQL_NULL. dict_col_t::def_val: The 'default row' value of the column. If the column is not added instantly, def_val.len will be UNIV_SQL_DEFAULT. dict_col_t: Add the accessors is_virtual(), is_nullable(), is_instant(), instant_value(). dict_col_t::remove_instant(): Remove the 'instant ADD' status of a column. dict_col_t::name(const dict_table_t& table): Replaces dict_table_get_col_name(). dict_index_t::n_core_fields: The original number of fields. For secondary indexes and if instant ADD COLUMN has not been used, this will be equal to dict_index_t::n_fields. dict_index_t::n_core_null_bytes: Number of bytes needed to represent the null flags; usually equal to UT_BITS_IN_BYTES(n_nullable). dict_index_t::NO_CORE_NULL_BYTES: Magic value signalling that n_core_null_bytes was not initialized yet from the clustered index root page. dict_index_t: Add the accessors is_instant(), is_clust(), get_n_nullable(), instant_field_value(). dict_index_t::instant_add_field(): Adjust clustered index metadata for instant ADD COLUMN. dict_index_t::remove_instant(): Remove the 'instant ADD' status of a clustered index when the table becomes empty, or the very first instant ADD COLUMN operation is rolled back. dict_table_t: Add the accessors is_instant(), is_temporary(), supports_instant(). dict_table_t::instant_add_column(): Adjust metadata for instant ADD COLUMN. dict_table_t::rollback_instant(): Adjust metadata on the rollback of instant ADD COLUMN. prepare_inplace_alter_table_dict(): First create the ctx->new_table, and only then decide if the table really needs to be rebuilt. We must split the creation of table or index metadata from the creation of the dictionary table records and the creation of the data. In this way, we can transform a table-rebuilding operation into an instant ADD COLUMN operation. Dictionary objects will only be added to cache when table rebuilding or index creation is needed. The ctx->instant_table will never be added to cache. dict_table_t::add_to_cache(): Modified and renamed from dict_table_add_to_cache(). Do not modify the table metadata. Let the callers invoke dict_table_add_system_columns() and if needed, set can_be_evicted. dict_create_sys_tables_tuple(), dict_create_table_step(): Omit the system columns (which will now exist in the dict_table_t object already at this point). dict_create_table_step(): Expect the callers to invoke dict_table_add_system_columns(). pars_create_table(): Before creating the table creation execution graph, invoke dict_table_add_system_columns(). row_create_table_for_mysql(): Expect all callers to invoke dict_table_add_system_columns(). create_index_dict(): Replaces row_merge_create_index_graph(). innodb_update_n_cols(): Renamed from innobase_update_n_virtual(). Call my_error() if an error occurs. btr_cur_instant_init(), btr_cur_instant_init_low(), btr_cur_instant_root_init(): Load additional metadata from the clustered index and set dict_index_t::n_core_null_bytes. This is invoked when table metadata is first loaded into the data dictionary. dict_boot(): Initialize n_core_null_bytes for the four hard-coded dictionary tables. dict_create_index_step(): Initialize n_core_null_bytes. This is executed as part of CREATE TABLE. dict_index_build_internal_clust(): Initialize n_core_null_bytes to NO_CORE_NULL_BYTES if table->supports_instant(). row_create_index_for_mysql(): Initialize n_core_null_bytes for CREATE TEMPORARY TABLE. commit_cache_norebuild(): Call the code to rename or enlarge columns in the cache only if instant ADD COLUMN is not being used. (Instant ADD COLUMN would copy all column metadata from instant_table to old_table, including the names and lengths.) PAGE_INSTANT: A new 13-bit field for storing dict_index_t::n_core_fields. This is repurposing the 16-bit field PAGE_DIRECTION, of which only the least significant 3 bits were used. The original byte containing PAGE_DIRECTION will be accessible via the new constant PAGE_DIRECTION_B. page_get_instant(), page_set_instant(): Accessors for the PAGE_INSTANT. page_ptr_get_direction(), page_get_direction(), page_ptr_set_direction(): Accessors for PAGE_DIRECTION. page_direction_reset(): Reset PAGE_DIRECTION, PAGE_N_DIRECTION. page_direction_increment(): Increment PAGE_N_DIRECTION and set PAGE_DIRECTION. rec_get_offsets(): Use the 'leaf' parameter for non-debug purposes, and assume that heap_no is always set. Initialize all dict_index_t::n_fields for ROW_FORMAT=REDUNDANT records, even if the record contains fewer fields. rec_offs_make_valid(): Add the parameter 'leaf'. rec_copy_prefix_to_dtuple(): Assert that the tuple is only built on the core fields. Instant ADD COLUMN only applies to the clustered index, and we should never build a search key that has more than the PRIMARY KEY and possibly DB_TRX_ID,DB_ROLL_PTR. All these columns are always present. dict_index_build_data_tuple(): Remove assertions that would be duplicated in rec_copy_prefix_to_dtuple(). rec_init_offsets(): Support ROW_FORMAT=REDUNDANT records whose number of fields is between n_core_fields and n_fields. cmp_rec_rec_with_match(): Implement the comparison between two MIN_REC_FLAG records. trx_t::in_rollback: Make the field available in non-debug builds. trx_start_for_ddl_low(): Remove dangerous error-tolerance. A dictionary transaction must be flagged as such before it has generated any undo log records. This is because trx_undo_assign_undo() will mark the transaction as a dictionary transaction in the undo log header right before the very first undo log record is being written. btr_index_rec_validate(): Account for instant ADD COLUMN row_undo_ins_remove_clust_rec(): On the rollback of an insert into SYS_COLUMNS, revert instant ADD COLUMN in the cache by removing the last column from the table and the clustered index. row_search_on_row_ref(), row_undo_mod_parse_undo_rec(), row_undo_mod(), trx_undo_update_rec_get_update(): Handle the 'default row' as a special case. dtuple_t::trim(index): Omit a redundant suffix of an index tuple right before insert or update. After instant ADD COLUMN, if the last fields of a clustered index tuple match the 'default row', there is no need to store them. While trimming the entry, we must hold a page latch, so that the table cannot be emptied and the 'default row' be deleted. btr_cur_optimistic_update(), btr_cur_pessimistic_update(), row_upd_clust_rec_by_insert(), row_ins_clust_index_entry_low(): Invoke dtuple_t::trim() if needed. row_ins_clust_index_entry(): Restore dtuple_t::n_fields after calling row_ins_clust_index_entry_low(). rec_get_converted_size(), rec_get_converted_size_comp(): Allow the number of fields to be between n_core_fields and n_fields. Do not support infimum,supremum. They are never supposed to be stored in dtuple_t, because page creation nowadays uses a lower-level method for initializing them. rec_convert_dtuple_to_rec_comp(): Assign the status bits based on the number of fields. btr_cur_trim(): In an update, trim the index entry as needed. For the 'default row', handle rollback specially. For user records, omit fields that match the 'default row'. btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete(): Skip locking and adaptive hash index for the 'default row'. row_log_table_apply_convert_mrec(): Replace 'default row' values if needed. In the temporary file that is applied by row_log_table_apply(), we must identify whether the records contain the extra header for instantly added columns. For now, we will allocate an additional byte for this for ROW_T_INSERT and ROW_T_UPDATE records when the source table has been subject to instant ADD COLUMN. The ROW_T_DELETE records are fine, as they will be converted and will only contain 'core' columns (PRIMARY KEY and some system columns) that are converted from dtuple_t. rec_get_converted_size_temp(), rec_init_offsets_temp(), rec_convert_dtuple_to_temp(): Add the parameter 'status'. REC_INFO_DEFAULT_ROW = REC_INFO_MIN_REC_FLAG | REC_STATUS_COLUMNS_ADDED: An info_bits constant for distinguishing the 'default row' record. rec_comp_status_t: An enum of the status bit values. rec_leaf_format: An enum that replaces the bool parameter of rec_init_offsets_comp_ordinary().
8 years ago
MDEV-15662 Instant DROP COLUMN or changing the order of columns Allow ADD COLUMN anywhere in a table, not only adding as the last column. Allow instant DROP COLUMN and instant changing the order of columns. The added columns will always be added last in clustered index records. In new records, instantly dropped columns will be stored as NULL or empty when possible. Information about dropped and reordered columns will be written in a metadata BLOB (mblob), which is stored before the first 'user' field in the hidden metadata record at the start of the clustered index. The presence of mblob is indicated by setting the delete-mark flag in the metadata record. The metadata BLOB stores the number of clustered index fields, followed by an array of column information for each field. For dropped columns, we store the NOT NULL flag, the fixed length, and for variable-length columns, whether the maximum length exceeded 255 bytes. For non-dropped columns, we store the column position. Unlike with MDEV-11369, when a table becomes empty, it cannot be converted back to the canonical format. The reason for this is that other threads may hold cached objects such as row_prebuilt_t::ins_node that could refer to dropped or reordered index fields. For instant DROP COLUMN and ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC, we must store the n_core_null_bytes in the root page, so that the chain of node pointer records can be followed in order to reach the leftmost leaf page where the metadata record is located. If the mblob is present, we will zero-initialize the strings "infimum" and "supremum" in the root page, and use the last byte of "supremum" for storing the number of null bytes (which are allocated but useless on node pointer pages). This is necessary for btr_cur_instant_init_metadata() to be able to navigate to the mblob. If the PRIMARY KEY contains any variable-length column and some nullable columns were instantly dropped, the dict_index_t::n_nullable in the data dictionary could be smaller than it actually is in the non-leaf pages. Because of this, the non-leaf pages could use more bytes for the null flags than the data dictionary expects, and we could be reading the lengths of the variable-length columns from the wrong offset, and thus reading the child page number from wrong place. This is the result of two design mistakes that involve unnecessary storage of data: First, it is nonsense to store any data fields for the leftmost node pointer records, because the comparisons would be resolved by the MIN_REC_FLAG alone. Second, there cannot be any null fields in the clustered index node pointer fields, but we nevertheless reserve space for all the null flags. Limitations (future work): MDEV-17459 Allow instant ALTER TABLE even if FULLTEXT INDEX exists MDEV-17468 Avoid table rebuild on operations on generated columns MDEV-17494 Refuse ALGORITHM=INSTANT when the row size is too large btr_page_reorganize_low(): Preserve any metadata in the root page. Call lock_move_reorganize_page() only after restoring the "infimum" and "supremum" records, to avoid a memcmp() assertion failure. dict_col_t::DROPPED: Magic value for dict_col_t::ind. dict_col_t::clear_instant(): Renamed from dict_col_t::remove_instant(). Do not assert that the column was instantly added, because we sometimes call this unconditionally for all columns. Convert an instantly added column to a "core column". The old name remove_instant() could be mistaken to refer to "instant DROP COLUMN". dict_col_t::is_added(): Rename from dict_col_t::is_instant(). dtype_t::metadata_blob_init(): Initialize the mblob data type. dtuple_t::is_metadata(), dtuple_t::is_alter_metadata(), upd_t::is_metadata(), upd_t::is_alter_metadata(): Check if info_bits refer to a metadata record. dict_table_t::instant: Metadata about dropped or reordered columns. dict_table_t::prepare_instant(): Prepare ha_innobase_inplace_ctx::instant_table for instant ALTER TABLE. innobase_instant_try() will pass this to dict_table_t::instant_column(). On rollback, dict_table_t::rollback_instant() will be called. dict_table_t::instant_column(): Renamed from instant_add_column(). Add the parameter col_map so that columns can be reordered. Copy and adjust v_cols[] as well. dict_table_t::find(): Find an old column based on a new column number. dict_table_t::serialise_columns(), dict_table_t::deserialise_columns(): Convert the mblob. dict_index_t::instant_metadata(): Create the metadata record for instant ALTER TABLE. Invoke dict_table_t::serialise_columns(). dict_index_t::reconstruct_fields(): Invoked by dict_table_t::deserialise_columns(). dict_index_t::clear_instant_alter(): Move the fields for the dropped columns to the end, and sort the surviving index fields in ascending order of column position. ha_innobase::check_if_supported_inplace_alter(): Do not allow adding a FTS_DOC_ID column if a hidden FTS_DOC_ID column exists due to FULLTEXT INDEX. (This always required ALGORITHM=COPY.) instant_alter_column_possible(): Add a parameter for InnoDB table, to check for additional conditions, such as the maximum number of index fields. ha_innobase_inplace_ctx::first_alter_pos: The first column whose position is affected by instant ADD, DROP, or changing the order of columns. innobase_build_col_map(): Skip added virtual columns. prepare_inplace_add_virtual(): Correctly compute num_to_add_vcol. Remove some unnecessary code. Note that the call to innodb_base_col_setup() should be executed later. commit_try_norebuild(): If ctx->is_instant(), let the virtual columns be added or dropped by innobase_instant_try(). innobase_instant_try(): Fill in a zero default value for the hidden column FTS_DOC_ID (to reduce the work needed in MDEV-17459). If any columns were dropped or reordered (or added not last), delete any SYS_COLUMNS records for the following columns, and insert SYS_COLUMNS records for all subsequent stored columns as well as for all virtual columns. If any virtual column is dropped, rewrite all virtual column metadata. Use a shortcut only for adding virtual columns. This is because innobase_drop_virtual_try() assumes that the dropped virtual columns still exist in ctx->old_table. innodb_update_cols(): Renamed from innodb_update_n_cols(). innobase_add_one_virtual(), innobase_insert_sys_virtual(): Change the return type to bool, and invoke my_error() when detecting an error. innodb_insert_sys_columns(): Insert a record into SYS_COLUMNS. Refactored from innobase_add_one_virtual() and innobase_instant_add_col(). innobase_instant_add_col(): Replace the parameter dfield with type. innobase_instant_drop_cols(): Drop matching columns from SYS_COLUMNS and all columns from SYS_VIRTUAL. innobase_add_virtual_try(), innobase_drop_virtual_try(): Let the caller invoke innodb_update_cols(). innobase_rename_column_try(): Skip dropped columns. commit_cache_norebuild(): Update table->fts->doc_col. dict_mem_table_col_rename_low(): Skip dropped columns. trx_undo_rec_get_partial_row(): Skip dropped columns. trx_undo_update_rec_get_update(): Handle the metadata BLOB correctly. trx_undo_page_report_modify(): Avoid out-of-bounds access to record fields. Log metadata records consistently. Apparently, the first fields of a clustered index may be updated in an update_undo vector when the index is ID_IND of SYS_FOREIGN, as part of renaming the table during ALTER TABLE. Normally, updates of the PRIMARY KEY should be logged as delete-mark and an insert. row_undo_mod_parse_undo_rec(), row_purge_parse_undo_rec(): Use trx_undo_metadata. row_undo_mod_clust_low(): On metadata rollback, roll back the root page too. row_undo_mod_clust(): Relax an assertion. The delete-mark flag was repurposed for ALTER TABLE metadata records. row_rec_to_index_entry_impl(): Add the template parameter mblob and the optional parameter info_bits for specifying the desired new info bits. For the metadata tuple, allow conversion between the original format (ADD COLUMN only) and the generic format (with hidden BLOB). Add the optional parameter "pad" to determine whether the tuple should be padded to the index fields (on ALTER TABLE it should), or whether it should remain at its original size (on rollback). row_build_index_entry_low(): Clean up the code, removing redundant variables and conditions. For instantly dropped columns, generate a dummy value that is NULL, the empty string, or a fixed length of NUL bytes, depending on the type of the dropped column. row_upd_clust_rec_by_insert_inherit_func(): On the update of PRIMARY KEY of a record that contained a dropped column whose value was stored externally, we will be inserting a dummy NULL or empty string value to the field of the dropped column. The externally stored column would eventually be dropped when purge removes the delete-marked record for the old PRIMARY KEY value. btr_index_rec_validate(): Recognize the metadata record. btr_discard_only_page_on_level(): Preserve the generic instant ALTER TABLE metadata. btr_set_instant(): Replaces page_set_instant(). This sets a clustered index root page to the appropriate format, or upgrades from the MDEV-11369 instant ADD COLUMN to generic ALTER TABLE format. btr_cur_instant_init_low(): Read and validate the metadata BLOB page before reconstructing the dictionary information based on it. btr_cur_instant_init_metadata(): Do not read any lengths from the metadata record header before reading the BLOB. At this point, we would not actually know how many nullable fields the metadata record contains. btr_cur_instant_root_init(): Initialize n_core_null_bytes in one of two possible ways. btr_cur_trim(): Handle the mblob record. row_metadata_to_tuple(): Convert a metadata record to a data tuple, based on the new info_bits of the metadata record. btr_cur_pessimistic_update(): Invoke row_metadata_to_tuple() if needed. Invoke dtuple_convert_big_rec() for metadata records if the record is too large, or if the mblob is not yet marked as externally stored. btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete(): When the last user record is deleted, do not delete the generic instant ALTER TABLE metadata record. Only delete MDEV-11369 instant ADD COLUMN metadata records. btr_cur_optimistic_insert(): Avoid unnecessary computation of rec_size. btr_pcur_store_position(): Allow a logically empty page to contain a metadata record for generic ALTER TABLE. REC_INFO_DEFAULT_ROW_ADD: Renamed from REC_INFO_DEFAULT_ROW. This is for the old instant ADD COLUMN (MDEV-11369) only. REC_INFO_DEFAULT_ROW_ALTER: The more generic metadata record, with additional information for dropped or reordered columns. rec_info_bits_valid(): Remove. The only case when this would fail is when the record is the generic ALTER TABLE metadata record. rec_is_alter_metadata(): Check if a record is the metadata record for instant ALTER TABLE (other than ADD COLUMN). NOTE: This function must not be invoked on node pointer records, because the delete-mark flag in those records may be set (it is garbage), and then a debug assertion could fail because index->is_instant() does not necessarily hold. rec_is_add_metadata(): Check if a record is MDEV-11369 ADD COLUMN metadata record (not more generic instant ALTER TABLE). rec_get_converted_size_comp_prefix_low(): Assume that the metadata field will be stored externally. In dtuple_convert_big_rec() during the rec_get_converted_size() call, it would not be there yet. rec_get_converted_size_comp(): Replace status,fields,n_fields with tuple. rec_init_offsets_comp_ordinary(), rec_get_converted_size_comp_prefix_low(), rec_convert_dtuple_to_rec_comp(): Add template<bool mblob = false>. With mblob=true, process a record with a metadata BLOB. rec_copy_prefix_to_buf(): Assert that no fields beyond the key and system columns are being copied. Exclude the metadata BLOB field. rec_convert_dtuple_to_metadata_comp(): Convert an alter metadata tuple into a record. row_upd_index_replace_metadata(): Apply an update vector to an alter_metadata tuple. row_log_allocate(): Replace dict_index_t::is_instant() with a more appropriate condition that ignores dict_table_t::instant. Only a table on which the MDEV-11369 ADD COLUMN was performed can "lose its instantness" when it becomes empty. After instant DROP COLUMN or reordering columns, we cannot simply convert the table to the canonical format, because the data dictionary cache and all possibly existing references to it from other client connection threads would have to be adjusted. row_quiesce_write_index_fields(): Do not crash when the table contains an instantly dropped column. Thanks to Thirunarayanan Balathandayuthapani for discussing the design and implementing an initial prototype of this. Thanks to Matthias Leich for testing.
7 years ago
MDEV-11369 Instant ADD COLUMN for InnoDB For InnoDB tables, adding, dropping and reordering columns has required a rebuild of the table and all its indexes. Since MySQL 5.6 (and MariaDB 10.0) this has been supported online (LOCK=NONE), allowing concurrent modification of the tables. This work revises the InnoDB ROW_FORMAT=REDUNDANT, ROW_FORMAT=COMPACT and ROW_FORMAT=DYNAMIC so that columns can be appended instantaneously, with only minor changes performed to the table structure. The counter innodb_instant_alter_column in INFORMATION_SCHEMA.GLOBAL_STATUS is incremented whenever a table rebuild operation is converted into an instant ADD COLUMN operation. ROW_FORMAT=COMPRESSED tables will not support instant ADD COLUMN. Some usability limitations will be addressed in subsequent work: MDEV-13134 Introduce ALTER TABLE attributes ALGORITHM=NOCOPY and ALGORITHM=INSTANT MDEV-14016 Allow instant ADD COLUMN, ADD INDEX, LOCK=NONE The format of the clustered index (PRIMARY KEY) is changed as follows: (1) The FIL_PAGE_TYPE of the root page will be FIL_PAGE_TYPE_INSTANT, and a new field PAGE_INSTANT will contain the original number of fields in the clustered index ('core' fields). If instant ADD COLUMN has not been used or the table becomes empty, or the very first instant ADD COLUMN operation is rolled back, the fields PAGE_INSTANT and FIL_PAGE_TYPE will be reset to 0 and FIL_PAGE_INDEX. (2) A special 'default row' record is inserted into the leftmost leaf, between the page infimum and the first user record. This record is distinguished by the REC_INFO_MIN_REC_FLAG, and it is otherwise in the same format as records that contain values for the instantly added columns. This 'default row' always has the same number of fields as the clustered index according to the table definition. The values of 'core' fields are to be ignored. For other fields, the 'default row' will contain the default values as they were during the ALTER TABLE statement. (If the column default values are changed later, those values will only be stored in the .frm file. The 'default row' will contain the original evaluated values, which must be the same for every row.) The 'default row' must be completely hidden from higher-level access routines. Assertions have been added to ensure that no 'default row' is ever present in the adaptive hash index or in locked records. The 'default row' is never delete-marked. (3) In clustered index leaf page records, the number of fields must reside between the number of 'core' fields (dict_index_t::n_core_fields introduced in this work) and dict_index_t::n_fields. If the number of fields is less than dict_index_t::n_fields, the missing fields are replaced with the column value of the 'default row'. Note: The number of fields in the record may shrink if some of the last instantly added columns are updated to the value that is in the 'default row'. The function btr_cur_trim() implements this 'compression' on update and rollback; dtuple::trim() implements it on insert. (4) In ROW_FORMAT=COMPACT and ROW_FORMAT=DYNAMIC records, the new status value REC_STATUS_COLUMNS_ADDED will indicate the presence of a new record header that will encode n_fields-n_core_fields-1 in 1 or 2 bytes. (In ROW_FORMAT=REDUNDANT records, the record header always explicitly encodes the number of fields.) We introduce the undo log record type TRX_UNDO_INSERT_DEFAULT for covering the insert of the 'default row' record when instant ADD COLUMN is used for the first time. Subsequent instant ADD COLUMN can use TRX_UNDO_UPD_EXIST_REC. This is joint work with Vin Chen (陈福荣) from Tencent. The design that was discussed in April 2017 would not have allowed import or export of data files, because instead of the 'default row' it would have introduced a data dictionary table. The test rpl.rpl_alter_instant is exactly as contributed in pull request #408. The test innodb.instant_alter is based on a contributed test. The redo log record format changes for ROW_FORMAT=DYNAMIC and ROW_FORMAT=COMPACT are as contributed. (With this change present, crash recovery from MariaDB 10.3.1 will fail in spectacular ways!) Also the semantics of higher-level redo log records that modify the PAGE_INSTANT field is changed. The redo log format version identifier was already changed to LOG_HEADER_FORMAT_CURRENT=103 in MariaDB 10.3.1. Everything else has been rewritten by me. Thanks to Elena Stepanova, the code has been tested extensively. When rolling back an instant ADD COLUMN operation, we must empty the PAGE_FREE list after deleting or shortening the 'default row' record, by calling either btr_page_empty() or btr_page_reorganize(). We must know the size of each entry in the PAGE_FREE list. If rollback left a freed copy of the 'default row' in the PAGE_FREE list, we would be unable to determine its size (if it is in ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC) because it would contain more fields than the rolled-back definition of the clustered index. UNIV_SQL_DEFAULT: A new special constant that designates an instantly added column that is not present in the clustered index record. len_is_stored(): Check if a length is an actual length. There are two magic length values: UNIV_SQL_DEFAULT, UNIV_SQL_NULL. dict_col_t::def_val: The 'default row' value of the column. If the column is not added instantly, def_val.len will be UNIV_SQL_DEFAULT. dict_col_t: Add the accessors is_virtual(), is_nullable(), is_instant(), instant_value(). dict_col_t::remove_instant(): Remove the 'instant ADD' status of a column. dict_col_t::name(const dict_table_t& table): Replaces dict_table_get_col_name(). dict_index_t::n_core_fields: The original number of fields. For secondary indexes and if instant ADD COLUMN has not been used, this will be equal to dict_index_t::n_fields. dict_index_t::n_core_null_bytes: Number of bytes needed to represent the null flags; usually equal to UT_BITS_IN_BYTES(n_nullable). dict_index_t::NO_CORE_NULL_BYTES: Magic value signalling that n_core_null_bytes was not initialized yet from the clustered index root page. dict_index_t: Add the accessors is_instant(), is_clust(), get_n_nullable(), instant_field_value(). dict_index_t::instant_add_field(): Adjust clustered index metadata for instant ADD COLUMN. dict_index_t::remove_instant(): Remove the 'instant ADD' status of a clustered index when the table becomes empty, or the very first instant ADD COLUMN operation is rolled back. dict_table_t: Add the accessors is_instant(), is_temporary(), supports_instant(). dict_table_t::instant_add_column(): Adjust metadata for instant ADD COLUMN. dict_table_t::rollback_instant(): Adjust metadata on the rollback of instant ADD COLUMN. prepare_inplace_alter_table_dict(): First create the ctx->new_table, and only then decide if the table really needs to be rebuilt. We must split the creation of table or index metadata from the creation of the dictionary table records and the creation of the data. In this way, we can transform a table-rebuilding operation into an instant ADD COLUMN operation. Dictionary objects will only be added to cache when table rebuilding or index creation is needed. The ctx->instant_table will never be added to cache. dict_table_t::add_to_cache(): Modified and renamed from dict_table_add_to_cache(). Do not modify the table metadata. Let the callers invoke dict_table_add_system_columns() and if needed, set can_be_evicted. dict_create_sys_tables_tuple(), dict_create_table_step(): Omit the system columns (which will now exist in the dict_table_t object already at this point). dict_create_table_step(): Expect the callers to invoke dict_table_add_system_columns(). pars_create_table(): Before creating the table creation execution graph, invoke dict_table_add_system_columns(). row_create_table_for_mysql(): Expect all callers to invoke dict_table_add_system_columns(). create_index_dict(): Replaces row_merge_create_index_graph(). innodb_update_n_cols(): Renamed from innobase_update_n_virtual(). Call my_error() if an error occurs. btr_cur_instant_init(), btr_cur_instant_init_low(), btr_cur_instant_root_init(): Load additional metadata from the clustered index and set dict_index_t::n_core_null_bytes. This is invoked when table metadata is first loaded into the data dictionary. dict_boot(): Initialize n_core_null_bytes for the four hard-coded dictionary tables. dict_create_index_step(): Initialize n_core_null_bytes. This is executed as part of CREATE TABLE. dict_index_build_internal_clust(): Initialize n_core_null_bytes to NO_CORE_NULL_BYTES if table->supports_instant(). row_create_index_for_mysql(): Initialize n_core_null_bytes for CREATE TEMPORARY TABLE. commit_cache_norebuild(): Call the code to rename or enlarge columns in the cache only if instant ADD COLUMN is not being used. (Instant ADD COLUMN would copy all column metadata from instant_table to old_table, including the names and lengths.) PAGE_INSTANT: A new 13-bit field for storing dict_index_t::n_core_fields. This is repurposing the 16-bit field PAGE_DIRECTION, of which only the least significant 3 bits were used. The original byte containing PAGE_DIRECTION will be accessible via the new constant PAGE_DIRECTION_B. page_get_instant(), page_set_instant(): Accessors for the PAGE_INSTANT. page_ptr_get_direction(), page_get_direction(), page_ptr_set_direction(): Accessors for PAGE_DIRECTION. page_direction_reset(): Reset PAGE_DIRECTION, PAGE_N_DIRECTION. page_direction_increment(): Increment PAGE_N_DIRECTION and set PAGE_DIRECTION. rec_get_offsets(): Use the 'leaf' parameter for non-debug purposes, and assume that heap_no is always set. Initialize all dict_index_t::n_fields for ROW_FORMAT=REDUNDANT records, even if the record contains fewer fields. rec_offs_make_valid(): Add the parameter 'leaf'. rec_copy_prefix_to_dtuple(): Assert that the tuple is only built on the core fields. Instant ADD COLUMN only applies to the clustered index, and we should never build a search key that has more than the PRIMARY KEY and possibly DB_TRX_ID,DB_ROLL_PTR. All these columns are always present. dict_index_build_data_tuple(): Remove assertions that would be duplicated in rec_copy_prefix_to_dtuple(). rec_init_offsets(): Support ROW_FORMAT=REDUNDANT records whose number of fields is between n_core_fields and n_fields. cmp_rec_rec_with_match(): Implement the comparison between two MIN_REC_FLAG records. trx_t::in_rollback: Make the field available in non-debug builds. trx_start_for_ddl_low(): Remove dangerous error-tolerance. A dictionary transaction must be flagged as such before it has generated any undo log records. This is because trx_undo_assign_undo() will mark the transaction as a dictionary transaction in the undo log header right before the very first undo log record is being written. btr_index_rec_validate(): Account for instant ADD COLUMN row_undo_ins_remove_clust_rec(): On the rollback of an insert into SYS_COLUMNS, revert instant ADD COLUMN in the cache by removing the last column from the table and the clustered index. row_search_on_row_ref(), row_undo_mod_parse_undo_rec(), row_undo_mod(), trx_undo_update_rec_get_update(): Handle the 'default row' as a special case. dtuple_t::trim(index): Omit a redundant suffix of an index tuple right before insert or update. After instant ADD COLUMN, if the last fields of a clustered index tuple match the 'default row', there is no need to store them. While trimming the entry, we must hold a page latch, so that the table cannot be emptied and the 'default row' be deleted. btr_cur_optimistic_update(), btr_cur_pessimistic_update(), row_upd_clust_rec_by_insert(), row_ins_clust_index_entry_low(): Invoke dtuple_t::trim() if needed. row_ins_clust_index_entry(): Restore dtuple_t::n_fields after calling row_ins_clust_index_entry_low(). rec_get_converted_size(), rec_get_converted_size_comp(): Allow the number of fields to be between n_core_fields and n_fields. Do not support infimum,supremum. They are never supposed to be stored in dtuple_t, because page creation nowadays uses a lower-level method for initializing them. rec_convert_dtuple_to_rec_comp(): Assign the status bits based on the number of fields. btr_cur_trim(): In an update, trim the index entry as needed. For the 'default row', handle rollback specially. For user records, omit fields that match the 'default row'. btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete(): Skip locking and adaptive hash index for the 'default row'. row_log_table_apply_convert_mrec(): Replace 'default row' values if needed. In the temporary file that is applied by row_log_table_apply(), we must identify whether the records contain the extra header for instantly added columns. For now, we will allocate an additional byte for this for ROW_T_INSERT and ROW_T_UPDATE records when the source table has been subject to instant ADD COLUMN. The ROW_T_DELETE records are fine, as they will be converted and will only contain 'core' columns (PRIMARY KEY and some system columns) that are converted from dtuple_t. rec_get_converted_size_temp(), rec_init_offsets_temp(), rec_convert_dtuple_to_temp(): Add the parameter 'status'. REC_INFO_DEFAULT_ROW = REC_INFO_MIN_REC_FLAG | REC_STATUS_COLUMNS_ADDED: An info_bits constant for distinguishing the 'default row' record. rec_comp_status_t: An enum of the status bit values. rec_leaf_format: An enum that replaces the bool parameter of rec_init_offsets_comp_ordinary().
8 years ago
MDEV-15662 Instant DROP COLUMN or changing the order of columns Allow ADD COLUMN anywhere in a table, not only adding as the last column. Allow instant DROP COLUMN and instant changing the order of columns. The added columns will always be added last in clustered index records. In new records, instantly dropped columns will be stored as NULL or empty when possible. Information about dropped and reordered columns will be written in a metadata BLOB (mblob), which is stored before the first 'user' field in the hidden metadata record at the start of the clustered index. The presence of mblob is indicated by setting the delete-mark flag in the metadata record. The metadata BLOB stores the number of clustered index fields, followed by an array of column information for each field. For dropped columns, we store the NOT NULL flag, the fixed length, and for variable-length columns, whether the maximum length exceeded 255 bytes. For non-dropped columns, we store the column position. Unlike with MDEV-11369, when a table becomes empty, it cannot be converted back to the canonical format. The reason for this is that other threads may hold cached objects such as row_prebuilt_t::ins_node that could refer to dropped or reordered index fields. For instant DROP COLUMN and ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC, we must store the n_core_null_bytes in the root page, so that the chain of node pointer records can be followed in order to reach the leftmost leaf page where the metadata record is located. If the mblob is present, we will zero-initialize the strings "infimum" and "supremum" in the root page, and use the last byte of "supremum" for storing the number of null bytes (which are allocated but useless on node pointer pages). This is necessary for btr_cur_instant_init_metadata() to be able to navigate to the mblob. If the PRIMARY KEY contains any variable-length column and some nullable columns were instantly dropped, the dict_index_t::n_nullable in the data dictionary could be smaller than it actually is in the non-leaf pages. Because of this, the non-leaf pages could use more bytes for the null flags than the data dictionary expects, and we could be reading the lengths of the variable-length columns from the wrong offset, and thus reading the child page number from wrong place. This is the result of two design mistakes that involve unnecessary storage of data: First, it is nonsense to store any data fields for the leftmost node pointer records, because the comparisons would be resolved by the MIN_REC_FLAG alone. Second, there cannot be any null fields in the clustered index node pointer fields, but we nevertheless reserve space for all the null flags. Limitations (future work): MDEV-17459 Allow instant ALTER TABLE even if FULLTEXT INDEX exists MDEV-17468 Avoid table rebuild on operations on generated columns MDEV-17494 Refuse ALGORITHM=INSTANT when the row size is too large btr_page_reorganize_low(): Preserve any metadata in the root page. Call lock_move_reorganize_page() only after restoring the "infimum" and "supremum" records, to avoid a memcmp() assertion failure. dict_col_t::DROPPED: Magic value for dict_col_t::ind. dict_col_t::clear_instant(): Renamed from dict_col_t::remove_instant(). Do not assert that the column was instantly added, because we sometimes call this unconditionally for all columns. Convert an instantly added column to a "core column". The old name remove_instant() could be mistaken to refer to "instant DROP COLUMN". dict_col_t::is_added(): Rename from dict_col_t::is_instant(). dtype_t::metadata_blob_init(): Initialize the mblob data type. dtuple_t::is_metadata(), dtuple_t::is_alter_metadata(), upd_t::is_metadata(), upd_t::is_alter_metadata(): Check if info_bits refer to a metadata record. dict_table_t::instant: Metadata about dropped or reordered columns. dict_table_t::prepare_instant(): Prepare ha_innobase_inplace_ctx::instant_table for instant ALTER TABLE. innobase_instant_try() will pass this to dict_table_t::instant_column(). On rollback, dict_table_t::rollback_instant() will be called. dict_table_t::instant_column(): Renamed from instant_add_column(). Add the parameter col_map so that columns can be reordered. Copy and adjust v_cols[] as well. dict_table_t::find(): Find an old column based on a new column number. dict_table_t::serialise_columns(), dict_table_t::deserialise_columns(): Convert the mblob. dict_index_t::instant_metadata(): Create the metadata record for instant ALTER TABLE. Invoke dict_table_t::serialise_columns(). dict_index_t::reconstruct_fields(): Invoked by dict_table_t::deserialise_columns(). dict_index_t::clear_instant_alter(): Move the fields for the dropped columns to the end, and sort the surviving index fields in ascending order of column position. ha_innobase::check_if_supported_inplace_alter(): Do not allow adding a FTS_DOC_ID column if a hidden FTS_DOC_ID column exists due to FULLTEXT INDEX. (This always required ALGORITHM=COPY.) instant_alter_column_possible(): Add a parameter for InnoDB table, to check for additional conditions, such as the maximum number of index fields. ha_innobase_inplace_ctx::first_alter_pos: The first column whose position is affected by instant ADD, DROP, or changing the order of columns. innobase_build_col_map(): Skip added virtual columns. prepare_inplace_add_virtual(): Correctly compute num_to_add_vcol. Remove some unnecessary code. Note that the call to innodb_base_col_setup() should be executed later. commit_try_norebuild(): If ctx->is_instant(), let the virtual columns be added or dropped by innobase_instant_try(). innobase_instant_try(): Fill in a zero default value for the hidden column FTS_DOC_ID (to reduce the work needed in MDEV-17459). If any columns were dropped or reordered (or added not last), delete any SYS_COLUMNS records for the following columns, and insert SYS_COLUMNS records for all subsequent stored columns as well as for all virtual columns. If any virtual column is dropped, rewrite all virtual column metadata. Use a shortcut only for adding virtual columns. This is because innobase_drop_virtual_try() assumes that the dropped virtual columns still exist in ctx->old_table. innodb_update_cols(): Renamed from innodb_update_n_cols(). innobase_add_one_virtual(), innobase_insert_sys_virtual(): Change the return type to bool, and invoke my_error() when detecting an error. innodb_insert_sys_columns(): Insert a record into SYS_COLUMNS. Refactored from innobase_add_one_virtual() and innobase_instant_add_col(). innobase_instant_add_col(): Replace the parameter dfield with type. innobase_instant_drop_cols(): Drop matching columns from SYS_COLUMNS and all columns from SYS_VIRTUAL. innobase_add_virtual_try(), innobase_drop_virtual_try(): Let the caller invoke innodb_update_cols(). innobase_rename_column_try(): Skip dropped columns. commit_cache_norebuild(): Update table->fts->doc_col. dict_mem_table_col_rename_low(): Skip dropped columns. trx_undo_rec_get_partial_row(): Skip dropped columns. trx_undo_update_rec_get_update(): Handle the metadata BLOB correctly. trx_undo_page_report_modify(): Avoid out-of-bounds access to record fields. Log metadata records consistently. Apparently, the first fields of a clustered index may be updated in an update_undo vector when the index is ID_IND of SYS_FOREIGN, as part of renaming the table during ALTER TABLE. Normally, updates of the PRIMARY KEY should be logged as delete-mark and an insert. row_undo_mod_parse_undo_rec(), row_purge_parse_undo_rec(): Use trx_undo_metadata. row_undo_mod_clust_low(): On metadata rollback, roll back the root page too. row_undo_mod_clust(): Relax an assertion. The delete-mark flag was repurposed for ALTER TABLE metadata records. row_rec_to_index_entry_impl(): Add the template parameter mblob and the optional parameter info_bits for specifying the desired new info bits. For the metadata tuple, allow conversion between the original format (ADD COLUMN only) and the generic format (with hidden BLOB). Add the optional parameter "pad" to determine whether the tuple should be padded to the index fields (on ALTER TABLE it should), or whether it should remain at its original size (on rollback). row_build_index_entry_low(): Clean up the code, removing redundant variables and conditions. For instantly dropped columns, generate a dummy value that is NULL, the empty string, or a fixed length of NUL bytes, depending on the type of the dropped column. row_upd_clust_rec_by_insert_inherit_func(): On the update of PRIMARY KEY of a record that contained a dropped column whose value was stored externally, we will be inserting a dummy NULL or empty string value to the field of the dropped column. The externally stored column would eventually be dropped when purge removes the delete-marked record for the old PRIMARY KEY value. btr_index_rec_validate(): Recognize the metadata record. btr_discard_only_page_on_level(): Preserve the generic instant ALTER TABLE metadata. btr_set_instant(): Replaces page_set_instant(). This sets a clustered index root page to the appropriate format, or upgrades from the MDEV-11369 instant ADD COLUMN to generic ALTER TABLE format. btr_cur_instant_init_low(): Read and validate the metadata BLOB page before reconstructing the dictionary information based on it. btr_cur_instant_init_metadata(): Do not read any lengths from the metadata record header before reading the BLOB. At this point, we would not actually know how many nullable fields the metadata record contains. btr_cur_instant_root_init(): Initialize n_core_null_bytes in one of two possible ways. btr_cur_trim(): Handle the mblob record. row_metadata_to_tuple(): Convert a metadata record to a data tuple, based on the new info_bits of the metadata record. btr_cur_pessimistic_update(): Invoke row_metadata_to_tuple() if needed. Invoke dtuple_convert_big_rec() for metadata records if the record is too large, or if the mblob is not yet marked as externally stored. btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete(): When the last user record is deleted, do not delete the generic instant ALTER TABLE metadata record. Only delete MDEV-11369 instant ADD COLUMN metadata records. btr_cur_optimistic_insert(): Avoid unnecessary computation of rec_size. btr_pcur_store_position(): Allow a logically empty page to contain a metadata record for generic ALTER TABLE. REC_INFO_DEFAULT_ROW_ADD: Renamed from REC_INFO_DEFAULT_ROW. This is for the old instant ADD COLUMN (MDEV-11369) only. REC_INFO_DEFAULT_ROW_ALTER: The more generic metadata record, with additional information for dropped or reordered columns. rec_info_bits_valid(): Remove. The only case when this would fail is when the record is the generic ALTER TABLE metadata record. rec_is_alter_metadata(): Check if a record is the metadata record for instant ALTER TABLE (other than ADD COLUMN). NOTE: This function must not be invoked on node pointer records, because the delete-mark flag in those records may be set (it is garbage), and then a debug assertion could fail because index->is_instant() does not necessarily hold. rec_is_add_metadata(): Check if a record is MDEV-11369 ADD COLUMN metadata record (not more generic instant ALTER TABLE). rec_get_converted_size_comp_prefix_low(): Assume that the metadata field will be stored externally. In dtuple_convert_big_rec() during the rec_get_converted_size() call, it would not be there yet. rec_get_converted_size_comp(): Replace status,fields,n_fields with tuple. rec_init_offsets_comp_ordinary(), rec_get_converted_size_comp_prefix_low(), rec_convert_dtuple_to_rec_comp(): Add template<bool mblob = false>. With mblob=true, process a record with a metadata BLOB. rec_copy_prefix_to_buf(): Assert that no fields beyond the key and system columns are being copied. Exclude the metadata BLOB field. rec_convert_dtuple_to_metadata_comp(): Convert an alter metadata tuple into a record. row_upd_index_replace_metadata(): Apply an update vector to an alter_metadata tuple. row_log_allocate(): Replace dict_index_t::is_instant() with a more appropriate condition that ignores dict_table_t::instant. Only a table on which the MDEV-11369 ADD COLUMN was performed can "lose its instantness" when it becomes empty. After instant DROP COLUMN or reordering columns, we cannot simply convert the table to the canonical format, because the data dictionary cache and all possibly existing references to it from other client connection threads would have to be adjusted. row_quiesce_write_index_fields(): Do not crash when the table contains an instantly dropped column. Thanks to Thirunarayanan Balathandayuthapani for discussing the design and implementing an initial prototype of this. Thanks to Matthias Leich for testing.
7 years ago
MDEV-11369 Instant ADD COLUMN for InnoDB For InnoDB tables, adding, dropping and reordering columns has required a rebuild of the table and all its indexes. Since MySQL 5.6 (and MariaDB 10.0) this has been supported online (LOCK=NONE), allowing concurrent modification of the tables. This work revises the InnoDB ROW_FORMAT=REDUNDANT, ROW_FORMAT=COMPACT and ROW_FORMAT=DYNAMIC so that columns can be appended instantaneously, with only minor changes performed to the table structure. The counter innodb_instant_alter_column in INFORMATION_SCHEMA.GLOBAL_STATUS is incremented whenever a table rebuild operation is converted into an instant ADD COLUMN operation. ROW_FORMAT=COMPRESSED tables will not support instant ADD COLUMN. Some usability limitations will be addressed in subsequent work: MDEV-13134 Introduce ALTER TABLE attributes ALGORITHM=NOCOPY and ALGORITHM=INSTANT MDEV-14016 Allow instant ADD COLUMN, ADD INDEX, LOCK=NONE The format of the clustered index (PRIMARY KEY) is changed as follows: (1) The FIL_PAGE_TYPE of the root page will be FIL_PAGE_TYPE_INSTANT, and a new field PAGE_INSTANT will contain the original number of fields in the clustered index ('core' fields). If instant ADD COLUMN has not been used or the table becomes empty, or the very first instant ADD COLUMN operation is rolled back, the fields PAGE_INSTANT and FIL_PAGE_TYPE will be reset to 0 and FIL_PAGE_INDEX. (2) A special 'default row' record is inserted into the leftmost leaf, between the page infimum and the first user record. This record is distinguished by the REC_INFO_MIN_REC_FLAG, and it is otherwise in the same format as records that contain values for the instantly added columns. This 'default row' always has the same number of fields as the clustered index according to the table definition. The values of 'core' fields are to be ignored. For other fields, the 'default row' will contain the default values as they were during the ALTER TABLE statement. (If the column default values are changed later, those values will only be stored in the .frm file. The 'default row' will contain the original evaluated values, which must be the same for every row.) The 'default row' must be completely hidden from higher-level access routines. Assertions have been added to ensure that no 'default row' is ever present in the adaptive hash index or in locked records. The 'default row' is never delete-marked. (3) In clustered index leaf page records, the number of fields must reside between the number of 'core' fields (dict_index_t::n_core_fields introduced in this work) and dict_index_t::n_fields. If the number of fields is less than dict_index_t::n_fields, the missing fields are replaced with the column value of the 'default row'. Note: The number of fields in the record may shrink if some of the last instantly added columns are updated to the value that is in the 'default row'. The function btr_cur_trim() implements this 'compression' on update and rollback; dtuple::trim() implements it on insert. (4) In ROW_FORMAT=COMPACT and ROW_FORMAT=DYNAMIC records, the new status value REC_STATUS_COLUMNS_ADDED will indicate the presence of a new record header that will encode n_fields-n_core_fields-1 in 1 or 2 bytes. (In ROW_FORMAT=REDUNDANT records, the record header always explicitly encodes the number of fields.) We introduce the undo log record type TRX_UNDO_INSERT_DEFAULT for covering the insert of the 'default row' record when instant ADD COLUMN is used for the first time. Subsequent instant ADD COLUMN can use TRX_UNDO_UPD_EXIST_REC. This is joint work with Vin Chen (陈福荣) from Tencent. The design that was discussed in April 2017 would not have allowed import or export of data files, because instead of the 'default row' it would have introduced a data dictionary table. The test rpl.rpl_alter_instant is exactly as contributed in pull request #408. The test innodb.instant_alter is based on a contributed test. The redo log record format changes for ROW_FORMAT=DYNAMIC and ROW_FORMAT=COMPACT are as contributed. (With this change present, crash recovery from MariaDB 10.3.1 will fail in spectacular ways!) Also the semantics of higher-level redo log records that modify the PAGE_INSTANT field is changed. The redo log format version identifier was already changed to LOG_HEADER_FORMAT_CURRENT=103 in MariaDB 10.3.1. Everything else has been rewritten by me. Thanks to Elena Stepanova, the code has been tested extensively. When rolling back an instant ADD COLUMN operation, we must empty the PAGE_FREE list after deleting or shortening the 'default row' record, by calling either btr_page_empty() or btr_page_reorganize(). We must know the size of each entry in the PAGE_FREE list. If rollback left a freed copy of the 'default row' in the PAGE_FREE list, we would be unable to determine its size (if it is in ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC) because it would contain more fields than the rolled-back definition of the clustered index. UNIV_SQL_DEFAULT: A new special constant that designates an instantly added column that is not present in the clustered index record. len_is_stored(): Check if a length is an actual length. There are two magic length values: UNIV_SQL_DEFAULT, UNIV_SQL_NULL. dict_col_t::def_val: The 'default row' value of the column. If the column is not added instantly, def_val.len will be UNIV_SQL_DEFAULT. dict_col_t: Add the accessors is_virtual(), is_nullable(), is_instant(), instant_value(). dict_col_t::remove_instant(): Remove the 'instant ADD' status of a column. dict_col_t::name(const dict_table_t& table): Replaces dict_table_get_col_name(). dict_index_t::n_core_fields: The original number of fields. For secondary indexes and if instant ADD COLUMN has not been used, this will be equal to dict_index_t::n_fields. dict_index_t::n_core_null_bytes: Number of bytes needed to represent the null flags; usually equal to UT_BITS_IN_BYTES(n_nullable). dict_index_t::NO_CORE_NULL_BYTES: Magic value signalling that n_core_null_bytes was not initialized yet from the clustered index root page. dict_index_t: Add the accessors is_instant(), is_clust(), get_n_nullable(), instant_field_value(). dict_index_t::instant_add_field(): Adjust clustered index metadata for instant ADD COLUMN. dict_index_t::remove_instant(): Remove the 'instant ADD' status of a clustered index when the table becomes empty, or the very first instant ADD COLUMN operation is rolled back. dict_table_t: Add the accessors is_instant(), is_temporary(), supports_instant(). dict_table_t::instant_add_column(): Adjust metadata for instant ADD COLUMN. dict_table_t::rollback_instant(): Adjust metadata on the rollback of instant ADD COLUMN. prepare_inplace_alter_table_dict(): First create the ctx->new_table, and only then decide if the table really needs to be rebuilt. We must split the creation of table or index metadata from the creation of the dictionary table records and the creation of the data. In this way, we can transform a table-rebuilding operation into an instant ADD COLUMN operation. Dictionary objects will only be added to cache when table rebuilding or index creation is needed. The ctx->instant_table will never be added to cache. dict_table_t::add_to_cache(): Modified and renamed from dict_table_add_to_cache(). Do not modify the table metadata. Let the callers invoke dict_table_add_system_columns() and if needed, set can_be_evicted. dict_create_sys_tables_tuple(), dict_create_table_step(): Omit the system columns (which will now exist in the dict_table_t object already at this point). dict_create_table_step(): Expect the callers to invoke dict_table_add_system_columns(). pars_create_table(): Before creating the table creation execution graph, invoke dict_table_add_system_columns(). row_create_table_for_mysql(): Expect all callers to invoke dict_table_add_system_columns(). create_index_dict(): Replaces row_merge_create_index_graph(). innodb_update_n_cols(): Renamed from innobase_update_n_virtual(). Call my_error() if an error occurs. btr_cur_instant_init(), btr_cur_instant_init_low(), btr_cur_instant_root_init(): Load additional metadata from the clustered index and set dict_index_t::n_core_null_bytes. This is invoked when table metadata is first loaded into the data dictionary. dict_boot(): Initialize n_core_null_bytes for the four hard-coded dictionary tables. dict_create_index_step(): Initialize n_core_null_bytes. This is executed as part of CREATE TABLE. dict_index_build_internal_clust(): Initialize n_core_null_bytes to NO_CORE_NULL_BYTES if table->supports_instant(). row_create_index_for_mysql(): Initialize n_core_null_bytes for CREATE TEMPORARY TABLE. commit_cache_norebuild(): Call the code to rename or enlarge columns in the cache only if instant ADD COLUMN is not being used. (Instant ADD COLUMN would copy all column metadata from instant_table to old_table, including the names and lengths.) PAGE_INSTANT: A new 13-bit field for storing dict_index_t::n_core_fields. This is repurposing the 16-bit field PAGE_DIRECTION, of which only the least significant 3 bits were used. The original byte containing PAGE_DIRECTION will be accessible via the new constant PAGE_DIRECTION_B. page_get_instant(), page_set_instant(): Accessors for the PAGE_INSTANT. page_ptr_get_direction(), page_get_direction(), page_ptr_set_direction(): Accessors for PAGE_DIRECTION. page_direction_reset(): Reset PAGE_DIRECTION, PAGE_N_DIRECTION. page_direction_increment(): Increment PAGE_N_DIRECTION and set PAGE_DIRECTION. rec_get_offsets(): Use the 'leaf' parameter for non-debug purposes, and assume that heap_no is always set. Initialize all dict_index_t::n_fields for ROW_FORMAT=REDUNDANT records, even if the record contains fewer fields. rec_offs_make_valid(): Add the parameter 'leaf'. rec_copy_prefix_to_dtuple(): Assert that the tuple is only built on the core fields. Instant ADD COLUMN only applies to the clustered index, and we should never build a search key that has more than the PRIMARY KEY and possibly DB_TRX_ID,DB_ROLL_PTR. All these columns are always present. dict_index_build_data_tuple(): Remove assertions that would be duplicated in rec_copy_prefix_to_dtuple(). rec_init_offsets(): Support ROW_FORMAT=REDUNDANT records whose number of fields is between n_core_fields and n_fields. cmp_rec_rec_with_match(): Implement the comparison between two MIN_REC_FLAG records. trx_t::in_rollback: Make the field available in non-debug builds. trx_start_for_ddl_low(): Remove dangerous error-tolerance. A dictionary transaction must be flagged as such before it has generated any undo log records. This is because trx_undo_assign_undo() will mark the transaction as a dictionary transaction in the undo log header right before the very first undo log record is being written. btr_index_rec_validate(): Account for instant ADD COLUMN row_undo_ins_remove_clust_rec(): On the rollback of an insert into SYS_COLUMNS, revert instant ADD COLUMN in the cache by removing the last column from the table and the clustered index. row_search_on_row_ref(), row_undo_mod_parse_undo_rec(), row_undo_mod(), trx_undo_update_rec_get_update(): Handle the 'default row' as a special case. dtuple_t::trim(index): Omit a redundant suffix of an index tuple right before insert or update. After instant ADD COLUMN, if the last fields of a clustered index tuple match the 'default row', there is no need to store them. While trimming the entry, we must hold a page latch, so that the table cannot be emptied and the 'default row' be deleted. btr_cur_optimistic_update(), btr_cur_pessimistic_update(), row_upd_clust_rec_by_insert(), row_ins_clust_index_entry_low(): Invoke dtuple_t::trim() if needed. row_ins_clust_index_entry(): Restore dtuple_t::n_fields after calling row_ins_clust_index_entry_low(). rec_get_converted_size(), rec_get_converted_size_comp(): Allow the number of fields to be between n_core_fields and n_fields. Do not support infimum,supremum. They are never supposed to be stored in dtuple_t, because page creation nowadays uses a lower-level method for initializing them. rec_convert_dtuple_to_rec_comp(): Assign the status bits based on the number of fields. btr_cur_trim(): In an update, trim the index entry as needed. For the 'default row', handle rollback specially. For user records, omit fields that match the 'default row'. btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete(): Skip locking and adaptive hash index for the 'default row'. row_log_table_apply_convert_mrec(): Replace 'default row' values if needed. In the temporary file that is applied by row_log_table_apply(), we must identify whether the records contain the extra header for instantly added columns. For now, we will allocate an additional byte for this for ROW_T_INSERT and ROW_T_UPDATE records when the source table has been subject to instant ADD COLUMN. The ROW_T_DELETE records are fine, as they will be converted and will only contain 'core' columns (PRIMARY KEY and some system columns) that are converted from dtuple_t. rec_get_converted_size_temp(), rec_init_offsets_temp(), rec_convert_dtuple_to_temp(): Add the parameter 'status'. REC_INFO_DEFAULT_ROW = REC_INFO_MIN_REC_FLAG | REC_STATUS_COLUMNS_ADDED: An info_bits constant for distinguishing the 'default row' record. rec_comp_status_t: An enum of the status bit values. rec_leaf_format: An enum that replaces the bool parameter of rec_init_offsets_comp_ordinary().
8 years ago
MDEV-15662 Instant DROP COLUMN or changing the order of columns Allow ADD COLUMN anywhere in a table, not only adding as the last column. Allow instant DROP COLUMN and instant changing the order of columns. The added columns will always be added last in clustered index records. In new records, instantly dropped columns will be stored as NULL or empty when possible. Information about dropped and reordered columns will be written in a metadata BLOB (mblob), which is stored before the first 'user' field in the hidden metadata record at the start of the clustered index. The presence of mblob is indicated by setting the delete-mark flag in the metadata record. The metadata BLOB stores the number of clustered index fields, followed by an array of column information for each field. For dropped columns, we store the NOT NULL flag, the fixed length, and for variable-length columns, whether the maximum length exceeded 255 bytes. For non-dropped columns, we store the column position. Unlike with MDEV-11369, when a table becomes empty, it cannot be converted back to the canonical format. The reason for this is that other threads may hold cached objects such as row_prebuilt_t::ins_node that could refer to dropped or reordered index fields. For instant DROP COLUMN and ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC, we must store the n_core_null_bytes in the root page, so that the chain of node pointer records can be followed in order to reach the leftmost leaf page where the metadata record is located. If the mblob is present, we will zero-initialize the strings "infimum" and "supremum" in the root page, and use the last byte of "supremum" for storing the number of null bytes (which are allocated but useless on node pointer pages). This is necessary for btr_cur_instant_init_metadata() to be able to navigate to the mblob. If the PRIMARY KEY contains any variable-length column and some nullable columns were instantly dropped, the dict_index_t::n_nullable in the data dictionary could be smaller than it actually is in the non-leaf pages. Because of this, the non-leaf pages could use more bytes for the null flags than the data dictionary expects, and we could be reading the lengths of the variable-length columns from the wrong offset, and thus reading the child page number from wrong place. This is the result of two design mistakes that involve unnecessary storage of data: First, it is nonsense to store any data fields for the leftmost node pointer records, because the comparisons would be resolved by the MIN_REC_FLAG alone. Second, there cannot be any null fields in the clustered index node pointer fields, but we nevertheless reserve space for all the null flags. Limitations (future work): MDEV-17459 Allow instant ALTER TABLE even if FULLTEXT INDEX exists MDEV-17468 Avoid table rebuild on operations on generated columns MDEV-17494 Refuse ALGORITHM=INSTANT when the row size is too large btr_page_reorganize_low(): Preserve any metadata in the root page. Call lock_move_reorganize_page() only after restoring the "infimum" and "supremum" records, to avoid a memcmp() assertion failure. dict_col_t::DROPPED: Magic value for dict_col_t::ind. dict_col_t::clear_instant(): Renamed from dict_col_t::remove_instant(). Do not assert that the column was instantly added, because we sometimes call this unconditionally for all columns. Convert an instantly added column to a "core column". The old name remove_instant() could be mistaken to refer to "instant DROP COLUMN". dict_col_t::is_added(): Rename from dict_col_t::is_instant(). dtype_t::metadata_blob_init(): Initialize the mblob data type. dtuple_t::is_metadata(), dtuple_t::is_alter_metadata(), upd_t::is_metadata(), upd_t::is_alter_metadata(): Check if info_bits refer to a metadata record. dict_table_t::instant: Metadata about dropped or reordered columns. dict_table_t::prepare_instant(): Prepare ha_innobase_inplace_ctx::instant_table for instant ALTER TABLE. innobase_instant_try() will pass this to dict_table_t::instant_column(). On rollback, dict_table_t::rollback_instant() will be called. dict_table_t::instant_column(): Renamed from instant_add_column(). Add the parameter col_map so that columns can be reordered. Copy and adjust v_cols[] as well. dict_table_t::find(): Find an old column based on a new column number. dict_table_t::serialise_columns(), dict_table_t::deserialise_columns(): Convert the mblob. dict_index_t::instant_metadata(): Create the metadata record for instant ALTER TABLE. Invoke dict_table_t::serialise_columns(). dict_index_t::reconstruct_fields(): Invoked by dict_table_t::deserialise_columns(). dict_index_t::clear_instant_alter(): Move the fields for the dropped columns to the end, and sort the surviving index fields in ascending order of column position. ha_innobase::check_if_supported_inplace_alter(): Do not allow adding a FTS_DOC_ID column if a hidden FTS_DOC_ID column exists due to FULLTEXT INDEX. (This always required ALGORITHM=COPY.) instant_alter_column_possible(): Add a parameter for InnoDB table, to check for additional conditions, such as the maximum number of index fields. ha_innobase_inplace_ctx::first_alter_pos: The first column whose position is affected by instant ADD, DROP, or changing the order of columns. innobase_build_col_map(): Skip added virtual columns. prepare_inplace_add_virtual(): Correctly compute num_to_add_vcol. Remove some unnecessary code. Note that the call to innodb_base_col_setup() should be executed later. commit_try_norebuild(): If ctx->is_instant(), let the virtual columns be added or dropped by innobase_instant_try(). innobase_instant_try(): Fill in a zero default value for the hidden column FTS_DOC_ID (to reduce the work needed in MDEV-17459). If any columns were dropped or reordered (or added not last), delete any SYS_COLUMNS records for the following columns, and insert SYS_COLUMNS records for all subsequent stored columns as well as for all virtual columns. If any virtual column is dropped, rewrite all virtual column metadata. Use a shortcut only for adding virtual columns. This is because innobase_drop_virtual_try() assumes that the dropped virtual columns still exist in ctx->old_table. innodb_update_cols(): Renamed from innodb_update_n_cols(). innobase_add_one_virtual(), innobase_insert_sys_virtual(): Change the return type to bool, and invoke my_error() when detecting an error. innodb_insert_sys_columns(): Insert a record into SYS_COLUMNS. Refactored from innobase_add_one_virtual() and innobase_instant_add_col(). innobase_instant_add_col(): Replace the parameter dfield with type. innobase_instant_drop_cols(): Drop matching columns from SYS_COLUMNS and all columns from SYS_VIRTUAL. innobase_add_virtual_try(), innobase_drop_virtual_try(): Let the caller invoke innodb_update_cols(). innobase_rename_column_try(): Skip dropped columns. commit_cache_norebuild(): Update table->fts->doc_col. dict_mem_table_col_rename_low(): Skip dropped columns. trx_undo_rec_get_partial_row(): Skip dropped columns. trx_undo_update_rec_get_update(): Handle the metadata BLOB correctly. trx_undo_page_report_modify(): Avoid out-of-bounds access to record fields. Log metadata records consistently. Apparently, the first fields of a clustered index may be updated in an update_undo vector when the index is ID_IND of SYS_FOREIGN, as part of renaming the table during ALTER TABLE. Normally, updates of the PRIMARY KEY should be logged as delete-mark and an insert. row_undo_mod_parse_undo_rec(), row_purge_parse_undo_rec(): Use trx_undo_metadata. row_undo_mod_clust_low(): On metadata rollback, roll back the root page too. row_undo_mod_clust(): Relax an assertion. The delete-mark flag was repurposed for ALTER TABLE metadata records. row_rec_to_index_entry_impl(): Add the template parameter mblob and the optional parameter info_bits for specifying the desired new info bits. For the metadata tuple, allow conversion between the original format (ADD COLUMN only) and the generic format (with hidden BLOB). Add the optional parameter "pad" to determine whether the tuple should be padded to the index fields (on ALTER TABLE it should), or whether it should remain at its original size (on rollback). row_build_index_entry_low(): Clean up the code, removing redundant variables and conditions. For instantly dropped columns, generate a dummy value that is NULL, the empty string, or a fixed length of NUL bytes, depending on the type of the dropped column. row_upd_clust_rec_by_insert_inherit_func(): On the update of PRIMARY KEY of a record that contained a dropped column whose value was stored externally, we will be inserting a dummy NULL or empty string value to the field of the dropped column. The externally stored column would eventually be dropped when purge removes the delete-marked record for the old PRIMARY KEY value. btr_index_rec_validate(): Recognize the metadata record. btr_discard_only_page_on_level(): Preserve the generic instant ALTER TABLE metadata. btr_set_instant(): Replaces page_set_instant(). This sets a clustered index root page to the appropriate format, or upgrades from the MDEV-11369 instant ADD COLUMN to generic ALTER TABLE format. btr_cur_instant_init_low(): Read and validate the metadata BLOB page before reconstructing the dictionary information based on it. btr_cur_instant_init_metadata(): Do not read any lengths from the metadata record header before reading the BLOB. At this point, we would not actually know how many nullable fields the metadata record contains. btr_cur_instant_root_init(): Initialize n_core_null_bytes in one of two possible ways. btr_cur_trim(): Handle the mblob record. row_metadata_to_tuple(): Convert a metadata record to a data tuple, based on the new info_bits of the metadata record. btr_cur_pessimistic_update(): Invoke row_metadata_to_tuple() if needed. Invoke dtuple_convert_big_rec() for metadata records if the record is too large, or if the mblob is not yet marked as externally stored. btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete(): When the last user record is deleted, do not delete the generic instant ALTER TABLE metadata record. Only delete MDEV-11369 instant ADD COLUMN metadata records. btr_cur_optimistic_insert(): Avoid unnecessary computation of rec_size. btr_pcur_store_position(): Allow a logically empty page to contain a metadata record for generic ALTER TABLE. REC_INFO_DEFAULT_ROW_ADD: Renamed from REC_INFO_DEFAULT_ROW. This is for the old instant ADD COLUMN (MDEV-11369) only. REC_INFO_DEFAULT_ROW_ALTER: The more generic metadata record, with additional information for dropped or reordered columns. rec_info_bits_valid(): Remove. The only case when this would fail is when the record is the generic ALTER TABLE metadata record. rec_is_alter_metadata(): Check if a record is the metadata record for instant ALTER TABLE (other than ADD COLUMN). NOTE: This function must not be invoked on node pointer records, because the delete-mark flag in those records may be set (it is garbage), and then a debug assertion could fail because index->is_instant() does not necessarily hold. rec_is_add_metadata(): Check if a record is MDEV-11369 ADD COLUMN metadata record (not more generic instant ALTER TABLE). rec_get_converted_size_comp_prefix_low(): Assume that the metadata field will be stored externally. In dtuple_convert_big_rec() during the rec_get_converted_size() call, it would not be there yet. rec_get_converted_size_comp(): Replace status,fields,n_fields with tuple. rec_init_offsets_comp_ordinary(), rec_get_converted_size_comp_prefix_low(), rec_convert_dtuple_to_rec_comp(): Add template<bool mblob = false>. With mblob=true, process a record with a metadata BLOB. rec_copy_prefix_to_buf(): Assert that no fields beyond the key and system columns are being copied. Exclude the metadata BLOB field. rec_convert_dtuple_to_metadata_comp(): Convert an alter metadata tuple into a record. row_upd_index_replace_metadata(): Apply an update vector to an alter_metadata tuple. row_log_allocate(): Replace dict_index_t::is_instant() with a more appropriate condition that ignores dict_table_t::instant. Only a table on which the MDEV-11369 ADD COLUMN was performed can "lose its instantness" when it becomes empty. After instant DROP COLUMN or reordering columns, we cannot simply convert the table to the canonical format, because the data dictionary cache and all possibly existing references to it from other client connection threads would have to be adjusted. row_quiesce_write_index_fields(): Do not crash when the table contains an instantly dropped column. Thanks to Thirunarayanan Balathandayuthapani for discussing the design and implementing an initial prototype of this. Thanks to Matthias Leich for testing.
7 years ago
MDEV-21907: InnoDB: Enable -Wconversion on clang and GCC The -Wconversion in GCC seems to be stricter than in clang. GCC at least since version 4.4.7 issues truncation warnings for assignments to bitfields, while clang 10 appears to only issue warnings when the sizes in bytes rounded to the nearest integer powers of 2 are different. Before GCC 10.0.0, -Wconversion required more casts and would not allow some operations, such as x<<=1 or x+=1 on a data type that is narrower than int. GCC 5 (but not GCC 4, GCC 6, or any later version) is complaining about x|=y even when x and y are compatible types that are narrower than int. Hence, we must rewrite some x|=y as x=static_cast<byte>(x|y) or similar, or we must disable -Wconversion. In GCC 6 and later, the warning for assigning wider to bitfields that are narrower than 8, 16, or 32 bits can be suppressed by applying a bitwise & with the exact bitmask of the bitfield. For older GCC, we must disable -Wconversion for GCC 4 or 5 in such cases. The bitwise negation operator appears to promote short integers to a wider type, and hence we must add explicit truncation casts around them. Microsoft Visual C does not allow a static_cast to truncate a constant, such as static_cast<byte>(1) truncating int. Hence, we will use the constructor-style cast byte(~1) for such cases. This has been tested at least with GCC 4.8.5, 5.4.0, 7.4.0, 9.2.1, 10.0.0, clang 9.0.1, 10.0.0, and MSVC 14.22.27905 (Microsoft Visual Studio 2019) on 64-bit and 32-bit targets (IA-32, AMD64, POWER 8, POWER 9, ARMv8).
6 years ago
MDEV-15662 Instant DROP COLUMN or changing the order of columns Allow ADD COLUMN anywhere in a table, not only adding as the last column. Allow instant DROP COLUMN and instant changing the order of columns. The added columns will always be added last in clustered index records. In new records, instantly dropped columns will be stored as NULL or empty when possible. Information about dropped and reordered columns will be written in a metadata BLOB (mblob), which is stored before the first 'user' field in the hidden metadata record at the start of the clustered index. The presence of mblob is indicated by setting the delete-mark flag in the metadata record. The metadata BLOB stores the number of clustered index fields, followed by an array of column information for each field. For dropped columns, we store the NOT NULL flag, the fixed length, and for variable-length columns, whether the maximum length exceeded 255 bytes. For non-dropped columns, we store the column position. Unlike with MDEV-11369, when a table becomes empty, it cannot be converted back to the canonical format. The reason for this is that other threads may hold cached objects such as row_prebuilt_t::ins_node that could refer to dropped or reordered index fields. For instant DROP COLUMN and ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC, we must store the n_core_null_bytes in the root page, so that the chain of node pointer records can be followed in order to reach the leftmost leaf page where the metadata record is located. If the mblob is present, we will zero-initialize the strings "infimum" and "supremum" in the root page, and use the last byte of "supremum" for storing the number of null bytes (which are allocated but useless on node pointer pages). This is necessary for btr_cur_instant_init_metadata() to be able to navigate to the mblob. If the PRIMARY KEY contains any variable-length column and some nullable columns were instantly dropped, the dict_index_t::n_nullable in the data dictionary could be smaller than it actually is in the non-leaf pages. Because of this, the non-leaf pages could use more bytes for the null flags than the data dictionary expects, and we could be reading the lengths of the variable-length columns from the wrong offset, and thus reading the child page number from wrong place. This is the result of two design mistakes that involve unnecessary storage of data: First, it is nonsense to store any data fields for the leftmost node pointer records, because the comparisons would be resolved by the MIN_REC_FLAG alone. Second, there cannot be any null fields in the clustered index node pointer fields, but we nevertheless reserve space for all the null flags. Limitations (future work): MDEV-17459 Allow instant ALTER TABLE even if FULLTEXT INDEX exists MDEV-17468 Avoid table rebuild on operations on generated columns MDEV-17494 Refuse ALGORITHM=INSTANT when the row size is too large btr_page_reorganize_low(): Preserve any metadata in the root page. Call lock_move_reorganize_page() only after restoring the "infimum" and "supremum" records, to avoid a memcmp() assertion failure. dict_col_t::DROPPED: Magic value for dict_col_t::ind. dict_col_t::clear_instant(): Renamed from dict_col_t::remove_instant(). Do not assert that the column was instantly added, because we sometimes call this unconditionally for all columns. Convert an instantly added column to a "core column". The old name remove_instant() could be mistaken to refer to "instant DROP COLUMN". dict_col_t::is_added(): Rename from dict_col_t::is_instant(). dtype_t::metadata_blob_init(): Initialize the mblob data type. dtuple_t::is_metadata(), dtuple_t::is_alter_metadata(), upd_t::is_metadata(), upd_t::is_alter_metadata(): Check if info_bits refer to a metadata record. dict_table_t::instant: Metadata about dropped or reordered columns. dict_table_t::prepare_instant(): Prepare ha_innobase_inplace_ctx::instant_table for instant ALTER TABLE. innobase_instant_try() will pass this to dict_table_t::instant_column(). On rollback, dict_table_t::rollback_instant() will be called. dict_table_t::instant_column(): Renamed from instant_add_column(). Add the parameter col_map so that columns can be reordered. Copy and adjust v_cols[] as well. dict_table_t::find(): Find an old column based on a new column number. dict_table_t::serialise_columns(), dict_table_t::deserialise_columns(): Convert the mblob. dict_index_t::instant_metadata(): Create the metadata record for instant ALTER TABLE. Invoke dict_table_t::serialise_columns(). dict_index_t::reconstruct_fields(): Invoked by dict_table_t::deserialise_columns(). dict_index_t::clear_instant_alter(): Move the fields for the dropped columns to the end, and sort the surviving index fields in ascending order of column position. ha_innobase::check_if_supported_inplace_alter(): Do not allow adding a FTS_DOC_ID column if a hidden FTS_DOC_ID column exists due to FULLTEXT INDEX. (This always required ALGORITHM=COPY.) instant_alter_column_possible(): Add a parameter for InnoDB table, to check for additional conditions, such as the maximum number of index fields. ha_innobase_inplace_ctx::first_alter_pos: The first column whose position is affected by instant ADD, DROP, or changing the order of columns. innobase_build_col_map(): Skip added virtual columns. prepare_inplace_add_virtual(): Correctly compute num_to_add_vcol. Remove some unnecessary code. Note that the call to innodb_base_col_setup() should be executed later. commit_try_norebuild(): If ctx->is_instant(), let the virtual columns be added or dropped by innobase_instant_try(). innobase_instant_try(): Fill in a zero default value for the hidden column FTS_DOC_ID (to reduce the work needed in MDEV-17459). If any columns were dropped or reordered (or added not last), delete any SYS_COLUMNS records for the following columns, and insert SYS_COLUMNS records for all subsequent stored columns as well as for all virtual columns. If any virtual column is dropped, rewrite all virtual column metadata. Use a shortcut only for adding virtual columns. This is because innobase_drop_virtual_try() assumes that the dropped virtual columns still exist in ctx->old_table. innodb_update_cols(): Renamed from innodb_update_n_cols(). innobase_add_one_virtual(), innobase_insert_sys_virtual(): Change the return type to bool, and invoke my_error() when detecting an error. innodb_insert_sys_columns(): Insert a record into SYS_COLUMNS. Refactored from innobase_add_one_virtual() and innobase_instant_add_col(). innobase_instant_add_col(): Replace the parameter dfield with type. innobase_instant_drop_cols(): Drop matching columns from SYS_COLUMNS and all columns from SYS_VIRTUAL. innobase_add_virtual_try(), innobase_drop_virtual_try(): Let the caller invoke innodb_update_cols(). innobase_rename_column_try(): Skip dropped columns. commit_cache_norebuild(): Update table->fts->doc_col. dict_mem_table_col_rename_low(): Skip dropped columns. trx_undo_rec_get_partial_row(): Skip dropped columns. trx_undo_update_rec_get_update(): Handle the metadata BLOB correctly. trx_undo_page_report_modify(): Avoid out-of-bounds access to record fields. Log metadata records consistently. Apparently, the first fields of a clustered index may be updated in an update_undo vector when the index is ID_IND of SYS_FOREIGN, as part of renaming the table during ALTER TABLE. Normally, updates of the PRIMARY KEY should be logged as delete-mark and an insert. row_undo_mod_parse_undo_rec(), row_purge_parse_undo_rec(): Use trx_undo_metadata. row_undo_mod_clust_low(): On metadata rollback, roll back the root page too. row_undo_mod_clust(): Relax an assertion. The delete-mark flag was repurposed for ALTER TABLE metadata records. row_rec_to_index_entry_impl(): Add the template parameter mblob and the optional parameter info_bits for specifying the desired new info bits. For the metadata tuple, allow conversion between the original format (ADD COLUMN only) and the generic format (with hidden BLOB). Add the optional parameter "pad" to determine whether the tuple should be padded to the index fields (on ALTER TABLE it should), or whether it should remain at its original size (on rollback). row_build_index_entry_low(): Clean up the code, removing redundant variables and conditions. For instantly dropped columns, generate a dummy value that is NULL, the empty string, or a fixed length of NUL bytes, depending on the type of the dropped column. row_upd_clust_rec_by_insert_inherit_func(): On the update of PRIMARY KEY of a record that contained a dropped column whose value was stored externally, we will be inserting a dummy NULL or empty string value to the field of the dropped column. The externally stored column would eventually be dropped when purge removes the delete-marked record for the old PRIMARY KEY value. btr_index_rec_validate(): Recognize the metadata record. btr_discard_only_page_on_level(): Preserve the generic instant ALTER TABLE metadata. btr_set_instant(): Replaces page_set_instant(). This sets a clustered index root page to the appropriate format, or upgrades from the MDEV-11369 instant ADD COLUMN to generic ALTER TABLE format. btr_cur_instant_init_low(): Read and validate the metadata BLOB page before reconstructing the dictionary information based on it. btr_cur_instant_init_metadata(): Do not read any lengths from the metadata record header before reading the BLOB. At this point, we would not actually know how many nullable fields the metadata record contains. btr_cur_instant_root_init(): Initialize n_core_null_bytes in one of two possible ways. btr_cur_trim(): Handle the mblob record. row_metadata_to_tuple(): Convert a metadata record to a data tuple, based on the new info_bits of the metadata record. btr_cur_pessimistic_update(): Invoke row_metadata_to_tuple() if needed. Invoke dtuple_convert_big_rec() for metadata records if the record is too large, or if the mblob is not yet marked as externally stored. btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete(): When the last user record is deleted, do not delete the generic instant ALTER TABLE metadata record. Only delete MDEV-11369 instant ADD COLUMN metadata records. btr_cur_optimistic_insert(): Avoid unnecessary computation of rec_size. btr_pcur_store_position(): Allow a logically empty page to contain a metadata record for generic ALTER TABLE. REC_INFO_DEFAULT_ROW_ADD: Renamed from REC_INFO_DEFAULT_ROW. This is for the old instant ADD COLUMN (MDEV-11369) only. REC_INFO_DEFAULT_ROW_ALTER: The more generic metadata record, with additional information for dropped or reordered columns. rec_info_bits_valid(): Remove. The only case when this would fail is when the record is the generic ALTER TABLE metadata record. rec_is_alter_metadata(): Check if a record is the metadata record for instant ALTER TABLE (other than ADD COLUMN). NOTE: This function must not be invoked on node pointer records, because the delete-mark flag in those records may be set (it is garbage), and then a debug assertion could fail because index->is_instant() does not necessarily hold. rec_is_add_metadata(): Check if a record is MDEV-11369 ADD COLUMN metadata record (not more generic instant ALTER TABLE). rec_get_converted_size_comp_prefix_low(): Assume that the metadata field will be stored externally. In dtuple_convert_big_rec() during the rec_get_converted_size() call, it would not be there yet. rec_get_converted_size_comp(): Replace status,fields,n_fields with tuple. rec_init_offsets_comp_ordinary(), rec_get_converted_size_comp_prefix_low(), rec_convert_dtuple_to_rec_comp(): Add template<bool mblob = false>. With mblob=true, process a record with a metadata BLOB. rec_copy_prefix_to_buf(): Assert that no fields beyond the key and system columns are being copied. Exclude the metadata BLOB field. rec_convert_dtuple_to_metadata_comp(): Convert an alter metadata tuple into a record. row_upd_index_replace_metadata(): Apply an update vector to an alter_metadata tuple. row_log_allocate(): Replace dict_index_t::is_instant() with a more appropriate condition that ignores dict_table_t::instant. Only a table on which the MDEV-11369 ADD COLUMN was performed can "lose its instantness" when it becomes empty. After instant DROP COLUMN or reordering columns, we cannot simply convert the table to the canonical format, because the data dictionary cache and all possibly existing references to it from other client connection threads would have to be adjusted. row_quiesce_write_index_fields(): Do not crash when the table contains an instantly dropped column. Thanks to Thirunarayanan Balathandayuthapani for discussing the design and implementing an initial prototype of this. Thanks to Matthias Leich for testing.
7 years ago
MDEV-15662 Instant DROP COLUMN or changing the order of columns Allow ADD COLUMN anywhere in a table, not only adding as the last column. Allow instant DROP COLUMN and instant changing the order of columns. The added columns will always be added last in clustered index records. In new records, instantly dropped columns will be stored as NULL or empty when possible. Information about dropped and reordered columns will be written in a metadata BLOB (mblob), which is stored before the first 'user' field in the hidden metadata record at the start of the clustered index. The presence of mblob is indicated by setting the delete-mark flag in the metadata record. The metadata BLOB stores the number of clustered index fields, followed by an array of column information for each field. For dropped columns, we store the NOT NULL flag, the fixed length, and for variable-length columns, whether the maximum length exceeded 255 bytes. For non-dropped columns, we store the column position. Unlike with MDEV-11369, when a table becomes empty, it cannot be converted back to the canonical format. The reason for this is that other threads may hold cached objects such as row_prebuilt_t::ins_node that could refer to dropped or reordered index fields. For instant DROP COLUMN and ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC, we must store the n_core_null_bytes in the root page, so that the chain of node pointer records can be followed in order to reach the leftmost leaf page where the metadata record is located. If the mblob is present, we will zero-initialize the strings "infimum" and "supremum" in the root page, and use the last byte of "supremum" for storing the number of null bytes (which are allocated but useless on node pointer pages). This is necessary for btr_cur_instant_init_metadata() to be able to navigate to the mblob. If the PRIMARY KEY contains any variable-length column and some nullable columns were instantly dropped, the dict_index_t::n_nullable in the data dictionary could be smaller than it actually is in the non-leaf pages. Because of this, the non-leaf pages could use more bytes for the null flags than the data dictionary expects, and we could be reading the lengths of the variable-length columns from the wrong offset, and thus reading the child page number from wrong place. This is the result of two design mistakes that involve unnecessary storage of data: First, it is nonsense to store any data fields for the leftmost node pointer records, because the comparisons would be resolved by the MIN_REC_FLAG alone. Second, there cannot be any null fields in the clustered index node pointer fields, but we nevertheless reserve space for all the null flags. Limitations (future work): MDEV-17459 Allow instant ALTER TABLE even if FULLTEXT INDEX exists MDEV-17468 Avoid table rebuild on operations on generated columns MDEV-17494 Refuse ALGORITHM=INSTANT when the row size is too large btr_page_reorganize_low(): Preserve any metadata in the root page. Call lock_move_reorganize_page() only after restoring the "infimum" and "supremum" records, to avoid a memcmp() assertion failure. dict_col_t::DROPPED: Magic value for dict_col_t::ind. dict_col_t::clear_instant(): Renamed from dict_col_t::remove_instant(). Do not assert that the column was instantly added, because we sometimes call this unconditionally for all columns. Convert an instantly added column to a "core column". The old name remove_instant() could be mistaken to refer to "instant DROP COLUMN". dict_col_t::is_added(): Rename from dict_col_t::is_instant(). dtype_t::metadata_blob_init(): Initialize the mblob data type. dtuple_t::is_metadata(), dtuple_t::is_alter_metadata(), upd_t::is_metadata(), upd_t::is_alter_metadata(): Check if info_bits refer to a metadata record. dict_table_t::instant: Metadata about dropped or reordered columns. dict_table_t::prepare_instant(): Prepare ha_innobase_inplace_ctx::instant_table for instant ALTER TABLE. innobase_instant_try() will pass this to dict_table_t::instant_column(). On rollback, dict_table_t::rollback_instant() will be called. dict_table_t::instant_column(): Renamed from instant_add_column(). Add the parameter col_map so that columns can be reordered. Copy and adjust v_cols[] as well. dict_table_t::find(): Find an old column based on a new column number. dict_table_t::serialise_columns(), dict_table_t::deserialise_columns(): Convert the mblob. dict_index_t::instant_metadata(): Create the metadata record for instant ALTER TABLE. Invoke dict_table_t::serialise_columns(). dict_index_t::reconstruct_fields(): Invoked by dict_table_t::deserialise_columns(). dict_index_t::clear_instant_alter(): Move the fields for the dropped columns to the end, and sort the surviving index fields in ascending order of column position. ha_innobase::check_if_supported_inplace_alter(): Do not allow adding a FTS_DOC_ID column if a hidden FTS_DOC_ID column exists due to FULLTEXT INDEX. (This always required ALGORITHM=COPY.) instant_alter_column_possible(): Add a parameter for InnoDB table, to check for additional conditions, such as the maximum number of index fields. ha_innobase_inplace_ctx::first_alter_pos: The first column whose position is affected by instant ADD, DROP, or changing the order of columns. innobase_build_col_map(): Skip added virtual columns. prepare_inplace_add_virtual(): Correctly compute num_to_add_vcol. Remove some unnecessary code. Note that the call to innodb_base_col_setup() should be executed later. commit_try_norebuild(): If ctx->is_instant(), let the virtual columns be added or dropped by innobase_instant_try(). innobase_instant_try(): Fill in a zero default value for the hidden column FTS_DOC_ID (to reduce the work needed in MDEV-17459). If any columns were dropped or reordered (or added not last), delete any SYS_COLUMNS records for the following columns, and insert SYS_COLUMNS records for all subsequent stored columns as well as for all virtual columns. If any virtual column is dropped, rewrite all virtual column metadata. Use a shortcut only for adding virtual columns. This is because innobase_drop_virtual_try() assumes that the dropped virtual columns still exist in ctx->old_table. innodb_update_cols(): Renamed from innodb_update_n_cols(). innobase_add_one_virtual(), innobase_insert_sys_virtual(): Change the return type to bool, and invoke my_error() when detecting an error. innodb_insert_sys_columns(): Insert a record into SYS_COLUMNS. Refactored from innobase_add_one_virtual() and innobase_instant_add_col(). innobase_instant_add_col(): Replace the parameter dfield with type. innobase_instant_drop_cols(): Drop matching columns from SYS_COLUMNS and all columns from SYS_VIRTUAL. innobase_add_virtual_try(), innobase_drop_virtual_try(): Let the caller invoke innodb_update_cols(). innobase_rename_column_try(): Skip dropped columns. commit_cache_norebuild(): Update table->fts->doc_col. dict_mem_table_col_rename_low(): Skip dropped columns. trx_undo_rec_get_partial_row(): Skip dropped columns. trx_undo_update_rec_get_update(): Handle the metadata BLOB correctly. trx_undo_page_report_modify(): Avoid out-of-bounds access to record fields. Log metadata records consistently. Apparently, the first fields of a clustered index may be updated in an update_undo vector when the index is ID_IND of SYS_FOREIGN, as part of renaming the table during ALTER TABLE. Normally, updates of the PRIMARY KEY should be logged as delete-mark and an insert. row_undo_mod_parse_undo_rec(), row_purge_parse_undo_rec(): Use trx_undo_metadata. row_undo_mod_clust_low(): On metadata rollback, roll back the root page too. row_undo_mod_clust(): Relax an assertion. The delete-mark flag was repurposed for ALTER TABLE metadata records. row_rec_to_index_entry_impl(): Add the template parameter mblob and the optional parameter info_bits for specifying the desired new info bits. For the metadata tuple, allow conversion between the original format (ADD COLUMN only) and the generic format (with hidden BLOB). Add the optional parameter "pad" to determine whether the tuple should be padded to the index fields (on ALTER TABLE it should), or whether it should remain at its original size (on rollback). row_build_index_entry_low(): Clean up the code, removing redundant variables and conditions. For instantly dropped columns, generate a dummy value that is NULL, the empty string, or a fixed length of NUL bytes, depending on the type of the dropped column. row_upd_clust_rec_by_insert_inherit_func(): On the update of PRIMARY KEY of a record that contained a dropped column whose value was stored externally, we will be inserting a dummy NULL or empty string value to the field of the dropped column. The externally stored column would eventually be dropped when purge removes the delete-marked record for the old PRIMARY KEY value. btr_index_rec_validate(): Recognize the metadata record. btr_discard_only_page_on_level(): Preserve the generic instant ALTER TABLE metadata. btr_set_instant(): Replaces page_set_instant(). This sets a clustered index root page to the appropriate format, or upgrades from the MDEV-11369 instant ADD COLUMN to generic ALTER TABLE format. btr_cur_instant_init_low(): Read and validate the metadata BLOB page before reconstructing the dictionary information based on it. btr_cur_instant_init_metadata(): Do not read any lengths from the metadata record header before reading the BLOB. At this point, we would not actually know how many nullable fields the metadata record contains. btr_cur_instant_root_init(): Initialize n_core_null_bytes in one of two possible ways. btr_cur_trim(): Handle the mblob record. row_metadata_to_tuple(): Convert a metadata record to a data tuple, based on the new info_bits of the metadata record. btr_cur_pessimistic_update(): Invoke row_metadata_to_tuple() if needed. Invoke dtuple_convert_big_rec() for metadata records if the record is too large, or if the mblob is not yet marked as externally stored. btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete(): When the last user record is deleted, do not delete the generic instant ALTER TABLE metadata record. Only delete MDEV-11369 instant ADD COLUMN metadata records. btr_cur_optimistic_insert(): Avoid unnecessary computation of rec_size. btr_pcur_store_position(): Allow a logically empty page to contain a metadata record for generic ALTER TABLE. REC_INFO_DEFAULT_ROW_ADD: Renamed from REC_INFO_DEFAULT_ROW. This is for the old instant ADD COLUMN (MDEV-11369) only. REC_INFO_DEFAULT_ROW_ALTER: The more generic metadata record, with additional information for dropped or reordered columns. rec_info_bits_valid(): Remove. The only case when this would fail is when the record is the generic ALTER TABLE metadata record. rec_is_alter_metadata(): Check if a record is the metadata record for instant ALTER TABLE (other than ADD COLUMN). NOTE: This function must not be invoked on node pointer records, because the delete-mark flag in those records may be set (it is garbage), and then a debug assertion could fail because index->is_instant() does not necessarily hold. rec_is_add_metadata(): Check if a record is MDEV-11369 ADD COLUMN metadata record (not more generic instant ALTER TABLE). rec_get_converted_size_comp_prefix_low(): Assume that the metadata field will be stored externally. In dtuple_convert_big_rec() during the rec_get_converted_size() call, it would not be there yet. rec_get_converted_size_comp(): Replace status,fields,n_fields with tuple. rec_init_offsets_comp_ordinary(), rec_get_converted_size_comp_prefix_low(), rec_convert_dtuple_to_rec_comp(): Add template<bool mblob = false>. With mblob=true, process a record with a metadata BLOB. rec_copy_prefix_to_buf(): Assert that no fields beyond the key and system columns are being copied. Exclude the metadata BLOB field. rec_convert_dtuple_to_metadata_comp(): Convert an alter metadata tuple into a record. row_upd_index_replace_metadata(): Apply an update vector to an alter_metadata tuple. row_log_allocate(): Replace dict_index_t::is_instant() with a more appropriate condition that ignores dict_table_t::instant. Only a table on which the MDEV-11369 ADD COLUMN was performed can "lose its instantness" when it becomes empty. After instant DROP COLUMN or reordering columns, we cannot simply convert the table to the canonical format, because the data dictionary cache and all possibly existing references to it from other client connection threads would have to be adjusted. row_quiesce_write_index_fields(): Do not crash when the table contains an instantly dropped column. Thanks to Thirunarayanan Balathandayuthapani for discussing the design and implementing an initial prototype of this. Thanks to Matthias Leich for testing.
7 years ago
MDEV-15662 Instant DROP COLUMN or changing the order of columns Allow ADD COLUMN anywhere in a table, not only adding as the last column. Allow instant DROP COLUMN and instant changing the order of columns. The added columns will always be added last in clustered index records. In new records, instantly dropped columns will be stored as NULL or empty when possible. Information about dropped and reordered columns will be written in a metadata BLOB (mblob), which is stored before the first 'user' field in the hidden metadata record at the start of the clustered index. The presence of mblob is indicated by setting the delete-mark flag in the metadata record. The metadata BLOB stores the number of clustered index fields, followed by an array of column information for each field. For dropped columns, we store the NOT NULL flag, the fixed length, and for variable-length columns, whether the maximum length exceeded 255 bytes. For non-dropped columns, we store the column position. Unlike with MDEV-11369, when a table becomes empty, it cannot be converted back to the canonical format. The reason for this is that other threads may hold cached objects such as row_prebuilt_t::ins_node that could refer to dropped or reordered index fields. For instant DROP COLUMN and ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC, we must store the n_core_null_bytes in the root page, so that the chain of node pointer records can be followed in order to reach the leftmost leaf page where the metadata record is located. If the mblob is present, we will zero-initialize the strings "infimum" and "supremum" in the root page, and use the last byte of "supremum" for storing the number of null bytes (which are allocated but useless on node pointer pages). This is necessary for btr_cur_instant_init_metadata() to be able to navigate to the mblob. If the PRIMARY KEY contains any variable-length column and some nullable columns were instantly dropped, the dict_index_t::n_nullable in the data dictionary could be smaller than it actually is in the non-leaf pages. Because of this, the non-leaf pages could use more bytes for the null flags than the data dictionary expects, and we could be reading the lengths of the variable-length columns from the wrong offset, and thus reading the child page number from wrong place. This is the result of two design mistakes that involve unnecessary storage of data: First, it is nonsense to store any data fields for the leftmost node pointer records, because the comparisons would be resolved by the MIN_REC_FLAG alone. Second, there cannot be any null fields in the clustered index node pointer fields, but we nevertheless reserve space for all the null flags. Limitations (future work): MDEV-17459 Allow instant ALTER TABLE even if FULLTEXT INDEX exists MDEV-17468 Avoid table rebuild on operations on generated columns MDEV-17494 Refuse ALGORITHM=INSTANT when the row size is too large btr_page_reorganize_low(): Preserve any metadata in the root page. Call lock_move_reorganize_page() only after restoring the "infimum" and "supremum" records, to avoid a memcmp() assertion failure. dict_col_t::DROPPED: Magic value for dict_col_t::ind. dict_col_t::clear_instant(): Renamed from dict_col_t::remove_instant(). Do not assert that the column was instantly added, because we sometimes call this unconditionally for all columns. Convert an instantly added column to a "core column". The old name remove_instant() could be mistaken to refer to "instant DROP COLUMN". dict_col_t::is_added(): Rename from dict_col_t::is_instant(). dtype_t::metadata_blob_init(): Initialize the mblob data type. dtuple_t::is_metadata(), dtuple_t::is_alter_metadata(), upd_t::is_metadata(), upd_t::is_alter_metadata(): Check if info_bits refer to a metadata record. dict_table_t::instant: Metadata about dropped or reordered columns. dict_table_t::prepare_instant(): Prepare ha_innobase_inplace_ctx::instant_table for instant ALTER TABLE. innobase_instant_try() will pass this to dict_table_t::instant_column(). On rollback, dict_table_t::rollback_instant() will be called. dict_table_t::instant_column(): Renamed from instant_add_column(). Add the parameter col_map so that columns can be reordered. Copy and adjust v_cols[] as well. dict_table_t::find(): Find an old column based on a new column number. dict_table_t::serialise_columns(), dict_table_t::deserialise_columns(): Convert the mblob. dict_index_t::instant_metadata(): Create the metadata record for instant ALTER TABLE. Invoke dict_table_t::serialise_columns(). dict_index_t::reconstruct_fields(): Invoked by dict_table_t::deserialise_columns(). dict_index_t::clear_instant_alter(): Move the fields for the dropped columns to the end, and sort the surviving index fields in ascending order of column position. ha_innobase::check_if_supported_inplace_alter(): Do not allow adding a FTS_DOC_ID column if a hidden FTS_DOC_ID column exists due to FULLTEXT INDEX. (This always required ALGORITHM=COPY.) instant_alter_column_possible(): Add a parameter for InnoDB table, to check for additional conditions, such as the maximum number of index fields. ha_innobase_inplace_ctx::first_alter_pos: The first column whose position is affected by instant ADD, DROP, or changing the order of columns. innobase_build_col_map(): Skip added virtual columns. prepare_inplace_add_virtual(): Correctly compute num_to_add_vcol. Remove some unnecessary code. Note that the call to innodb_base_col_setup() should be executed later. commit_try_norebuild(): If ctx->is_instant(), let the virtual columns be added or dropped by innobase_instant_try(). innobase_instant_try(): Fill in a zero default value for the hidden column FTS_DOC_ID (to reduce the work needed in MDEV-17459). If any columns were dropped or reordered (or added not last), delete any SYS_COLUMNS records for the following columns, and insert SYS_COLUMNS records for all subsequent stored columns as well as for all virtual columns. If any virtual column is dropped, rewrite all virtual column metadata. Use a shortcut only for adding virtual columns. This is because innobase_drop_virtual_try() assumes that the dropped virtual columns still exist in ctx->old_table. innodb_update_cols(): Renamed from innodb_update_n_cols(). innobase_add_one_virtual(), innobase_insert_sys_virtual(): Change the return type to bool, and invoke my_error() when detecting an error. innodb_insert_sys_columns(): Insert a record into SYS_COLUMNS. Refactored from innobase_add_one_virtual() and innobase_instant_add_col(). innobase_instant_add_col(): Replace the parameter dfield with type. innobase_instant_drop_cols(): Drop matching columns from SYS_COLUMNS and all columns from SYS_VIRTUAL. innobase_add_virtual_try(), innobase_drop_virtual_try(): Let the caller invoke innodb_update_cols(). innobase_rename_column_try(): Skip dropped columns. commit_cache_norebuild(): Update table->fts->doc_col. dict_mem_table_col_rename_low(): Skip dropped columns. trx_undo_rec_get_partial_row(): Skip dropped columns. trx_undo_update_rec_get_update(): Handle the metadata BLOB correctly. trx_undo_page_report_modify(): Avoid out-of-bounds access to record fields. Log metadata records consistently. Apparently, the first fields of a clustered index may be updated in an update_undo vector when the index is ID_IND of SYS_FOREIGN, as part of renaming the table during ALTER TABLE. Normally, updates of the PRIMARY KEY should be logged as delete-mark and an insert. row_undo_mod_parse_undo_rec(), row_purge_parse_undo_rec(): Use trx_undo_metadata. row_undo_mod_clust_low(): On metadata rollback, roll back the root page too. row_undo_mod_clust(): Relax an assertion. The delete-mark flag was repurposed for ALTER TABLE metadata records. row_rec_to_index_entry_impl(): Add the template parameter mblob and the optional parameter info_bits for specifying the desired new info bits. For the metadata tuple, allow conversion between the original format (ADD COLUMN only) and the generic format (with hidden BLOB). Add the optional parameter "pad" to determine whether the tuple should be padded to the index fields (on ALTER TABLE it should), or whether it should remain at its original size (on rollback). row_build_index_entry_low(): Clean up the code, removing redundant variables and conditions. For instantly dropped columns, generate a dummy value that is NULL, the empty string, or a fixed length of NUL bytes, depending on the type of the dropped column. row_upd_clust_rec_by_insert_inherit_func(): On the update of PRIMARY KEY of a record that contained a dropped column whose value was stored externally, we will be inserting a dummy NULL or empty string value to the field of the dropped column. The externally stored column would eventually be dropped when purge removes the delete-marked record for the old PRIMARY KEY value. btr_index_rec_validate(): Recognize the metadata record. btr_discard_only_page_on_level(): Preserve the generic instant ALTER TABLE metadata. btr_set_instant(): Replaces page_set_instant(). This sets a clustered index root page to the appropriate format, or upgrades from the MDEV-11369 instant ADD COLUMN to generic ALTER TABLE format. btr_cur_instant_init_low(): Read and validate the metadata BLOB page before reconstructing the dictionary information based on it. btr_cur_instant_init_metadata(): Do not read any lengths from the metadata record header before reading the BLOB. At this point, we would not actually know how many nullable fields the metadata record contains. btr_cur_instant_root_init(): Initialize n_core_null_bytes in one of two possible ways. btr_cur_trim(): Handle the mblob record. row_metadata_to_tuple(): Convert a metadata record to a data tuple, based on the new info_bits of the metadata record. btr_cur_pessimistic_update(): Invoke row_metadata_to_tuple() if needed. Invoke dtuple_convert_big_rec() for metadata records if the record is too large, or if the mblob is not yet marked as externally stored. btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete(): When the last user record is deleted, do not delete the generic instant ALTER TABLE metadata record. Only delete MDEV-11369 instant ADD COLUMN metadata records. btr_cur_optimistic_insert(): Avoid unnecessary computation of rec_size. btr_pcur_store_position(): Allow a logically empty page to contain a metadata record for generic ALTER TABLE. REC_INFO_DEFAULT_ROW_ADD: Renamed from REC_INFO_DEFAULT_ROW. This is for the old instant ADD COLUMN (MDEV-11369) only. REC_INFO_DEFAULT_ROW_ALTER: The more generic metadata record, with additional information for dropped or reordered columns. rec_info_bits_valid(): Remove. The only case when this would fail is when the record is the generic ALTER TABLE metadata record. rec_is_alter_metadata(): Check if a record is the metadata record for instant ALTER TABLE (other than ADD COLUMN). NOTE: This function must not be invoked on node pointer records, because the delete-mark flag in those records may be set (it is garbage), and then a debug assertion could fail because index->is_instant() does not necessarily hold. rec_is_add_metadata(): Check if a record is MDEV-11369 ADD COLUMN metadata record (not more generic instant ALTER TABLE). rec_get_converted_size_comp_prefix_low(): Assume that the metadata field will be stored externally. In dtuple_convert_big_rec() during the rec_get_converted_size() call, it would not be there yet. rec_get_converted_size_comp(): Replace status,fields,n_fields with tuple. rec_init_offsets_comp_ordinary(), rec_get_converted_size_comp_prefix_low(), rec_convert_dtuple_to_rec_comp(): Add template<bool mblob = false>. With mblob=true, process a record with a metadata BLOB. rec_copy_prefix_to_buf(): Assert that no fields beyond the key and system columns are being copied. Exclude the metadata BLOB field. rec_convert_dtuple_to_metadata_comp(): Convert an alter metadata tuple into a record. row_upd_index_replace_metadata(): Apply an update vector to an alter_metadata tuple. row_log_allocate(): Replace dict_index_t::is_instant() with a more appropriate condition that ignores dict_table_t::instant. Only a table on which the MDEV-11369 ADD COLUMN was performed can "lose its instantness" when it becomes empty. After instant DROP COLUMN or reordering columns, we cannot simply convert the table to the canonical format, because the data dictionary cache and all possibly existing references to it from other client connection threads would have to be adjusted. row_quiesce_write_index_fields(): Do not crash when the table contains an instantly dropped column. Thanks to Thirunarayanan Balathandayuthapani for discussing the design and implementing an initial prototype of this. Thanks to Matthias Leich for testing.
7 years ago
MDEV-15662 Instant DROP COLUMN or changing the order of columns Allow ADD COLUMN anywhere in a table, not only adding as the last column. Allow instant DROP COLUMN and instant changing the order of columns. The added columns will always be added last in clustered index records. In new records, instantly dropped columns will be stored as NULL or empty when possible. Information about dropped and reordered columns will be written in a metadata BLOB (mblob), which is stored before the first 'user' field in the hidden metadata record at the start of the clustered index. The presence of mblob is indicated by setting the delete-mark flag in the metadata record. The metadata BLOB stores the number of clustered index fields, followed by an array of column information for each field. For dropped columns, we store the NOT NULL flag, the fixed length, and for variable-length columns, whether the maximum length exceeded 255 bytes. For non-dropped columns, we store the column position. Unlike with MDEV-11369, when a table becomes empty, it cannot be converted back to the canonical format. The reason for this is that other threads may hold cached objects such as row_prebuilt_t::ins_node that could refer to dropped or reordered index fields. For instant DROP COLUMN and ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC, we must store the n_core_null_bytes in the root page, so that the chain of node pointer records can be followed in order to reach the leftmost leaf page where the metadata record is located. If the mblob is present, we will zero-initialize the strings "infimum" and "supremum" in the root page, and use the last byte of "supremum" for storing the number of null bytes (which are allocated but useless on node pointer pages). This is necessary for btr_cur_instant_init_metadata() to be able to navigate to the mblob. If the PRIMARY KEY contains any variable-length column and some nullable columns were instantly dropped, the dict_index_t::n_nullable in the data dictionary could be smaller than it actually is in the non-leaf pages. Because of this, the non-leaf pages could use more bytes for the null flags than the data dictionary expects, and we could be reading the lengths of the variable-length columns from the wrong offset, and thus reading the child page number from wrong place. This is the result of two design mistakes that involve unnecessary storage of data: First, it is nonsense to store any data fields for the leftmost node pointer records, because the comparisons would be resolved by the MIN_REC_FLAG alone. Second, there cannot be any null fields in the clustered index node pointer fields, but we nevertheless reserve space for all the null flags. Limitations (future work): MDEV-17459 Allow instant ALTER TABLE even if FULLTEXT INDEX exists MDEV-17468 Avoid table rebuild on operations on generated columns MDEV-17494 Refuse ALGORITHM=INSTANT when the row size is too large btr_page_reorganize_low(): Preserve any metadata in the root page. Call lock_move_reorganize_page() only after restoring the "infimum" and "supremum" records, to avoid a memcmp() assertion failure. dict_col_t::DROPPED: Magic value for dict_col_t::ind. dict_col_t::clear_instant(): Renamed from dict_col_t::remove_instant(). Do not assert that the column was instantly added, because we sometimes call this unconditionally for all columns. Convert an instantly added column to a "core column". The old name remove_instant() could be mistaken to refer to "instant DROP COLUMN". dict_col_t::is_added(): Rename from dict_col_t::is_instant(). dtype_t::metadata_blob_init(): Initialize the mblob data type. dtuple_t::is_metadata(), dtuple_t::is_alter_metadata(), upd_t::is_metadata(), upd_t::is_alter_metadata(): Check if info_bits refer to a metadata record. dict_table_t::instant: Metadata about dropped or reordered columns. dict_table_t::prepare_instant(): Prepare ha_innobase_inplace_ctx::instant_table for instant ALTER TABLE. innobase_instant_try() will pass this to dict_table_t::instant_column(). On rollback, dict_table_t::rollback_instant() will be called. dict_table_t::instant_column(): Renamed from instant_add_column(). Add the parameter col_map so that columns can be reordered. Copy and adjust v_cols[] as well. dict_table_t::find(): Find an old column based on a new column number. dict_table_t::serialise_columns(), dict_table_t::deserialise_columns(): Convert the mblob. dict_index_t::instant_metadata(): Create the metadata record for instant ALTER TABLE. Invoke dict_table_t::serialise_columns(). dict_index_t::reconstruct_fields(): Invoked by dict_table_t::deserialise_columns(). dict_index_t::clear_instant_alter(): Move the fields for the dropped columns to the end, and sort the surviving index fields in ascending order of column position. ha_innobase::check_if_supported_inplace_alter(): Do not allow adding a FTS_DOC_ID column if a hidden FTS_DOC_ID column exists due to FULLTEXT INDEX. (This always required ALGORITHM=COPY.) instant_alter_column_possible(): Add a parameter for InnoDB table, to check for additional conditions, such as the maximum number of index fields. ha_innobase_inplace_ctx::first_alter_pos: The first column whose position is affected by instant ADD, DROP, or changing the order of columns. innobase_build_col_map(): Skip added virtual columns. prepare_inplace_add_virtual(): Correctly compute num_to_add_vcol. Remove some unnecessary code. Note that the call to innodb_base_col_setup() should be executed later. commit_try_norebuild(): If ctx->is_instant(), let the virtual columns be added or dropped by innobase_instant_try(). innobase_instant_try(): Fill in a zero default value for the hidden column FTS_DOC_ID (to reduce the work needed in MDEV-17459). If any columns were dropped or reordered (or added not last), delete any SYS_COLUMNS records for the following columns, and insert SYS_COLUMNS records for all subsequent stored columns as well as for all virtual columns. If any virtual column is dropped, rewrite all virtual column metadata. Use a shortcut only for adding virtual columns. This is because innobase_drop_virtual_try() assumes that the dropped virtual columns still exist in ctx->old_table. innodb_update_cols(): Renamed from innodb_update_n_cols(). innobase_add_one_virtual(), innobase_insert_sys_virtual(): Change the return type to bool, and invoke my_error() when detecting an error. innodb_insert_sys_columns(): Insert a record into SYS_COLUMNS. Refactored from innobase_add_one_virtual() and innobase_instant_add_col(). innobase_instant_add_col(): Replace the parameter dfield with type. innobase_instant_drop_cols(): Drop matching columns from SYS_COLUMNS and all columns from SYS_VIRTUAL. innobase_add_virtual_try(), innobase_drop_virtual_try(): Let the caller invoke innodb_update_cols(). innobase_rename_column_try(): Skip dropped columns. commit_cache_norebuild(): Update table->fts->doc_col. dict_mem_table_col_rename_low(): Skip dropped columns. trx_undo_rec_get_partial_row(): Skip dropped columns. trx_undo_update_rec_get_update(): Handle the metadata BLOB correctly. trx_undo_page_report_modify(): Avoid out-of-bounds access to record fields. Log metadata records consistently. Apparently, the first fields of a clustered index may be updated in an update_undo vector when the index is ID_IND of SYS_FOREIGN, as part of renaming the table during ALTER TABLE. Normally, updates of the PRIMARY KEY should be logged as delete-mark and an insert. row_undo_mod_parse_undo_rec(), row_purge_parse_undo_rec(): Use trx_undo_metadata. row_undo_mod_clust_low(): On metadata rollback, roll back the root page too. row_undo_mod_clust(): Relax an assertion. The delete-mark flag was repurposed for ALTER TABLE metadata records. row_rec_to_index_entry_impl(): Add the template parameter mblob and the optional parameter info_bits for specifying the desired new info bits. For the metadata tuple, allow conversion between the original format (ADD COLUMN only) and the generic format (with hidden BLOB). Add the optional parameter "pad" to determine whether the tuple should be padded to the index fields (on ALTER TABLE it should), or whether it should remain at its original size (on rollback). row_build_index_entry_low(): Clean up the code, removing redundant variables and conditions. For instantly dropped columns, generate a dummy value that is NULL, the empty string, or a fixed length of NUL bytes, depending on the type of the dropped column. row_upd_clust_rec_by_insert_inherit_func(): On the update of PRIMARY KEY of a record that contained a dropped column whose value was stored externally, we will be inserting a dummy NULL or empty string value to the field of the dropped column. The externally stored column would eventually be dropped when purge removes the delete-marked record for the old PRIMARY KEY value. btr_index_rec_validate(): Recognize the metadata record. btr_discard_only_page_on_level(): Preserve the generic instant ALTER TABLE metadata. btr_set_instant(): Replaces page_set_instant(). This sets a clustered index root page to the appropriate format, or upgrades from the MDEV-11369 instant ADD COLUMN to generic ALTER TABLE format. btr_cur_instant_init_low(): Read and validate the metadata BLOB page before reconstructing the dictionary information based on it. btr_cur_instant_init_metadata(): Do not read any lengths from the metadata record header before reading the BLOB. At this point, we would not actually know how many nullable fields the metadata record contains. btr_cur_instant_root_init(): Initialize n_core_null_bytes in one of two possible ways. btr_cur_trim(): Handle the mblob record. row_metadata_to_tuple(): Convert a metadata record to a data tuple, based on the new info_bits of the metadata record. btr_cur_pessimistic_update(): Invoke row_metadata_to_tuple() if needed. Invoke dtuple_convert_big_rec() for metadata records if the record is too large, or if the mblob is not yet marked as externally stored. btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete(): When the last user record is deleted, do not delete the generic instant ALTER TABLE metadata record. Only delete MDEV-11369 instant ADD COLUMN metadata records. btr_cur_optimistic_insert(): Avoid unnecessary computation of rec_size. btr_pcur_store_position(): Allow a logically empty page to contain a metadata record for generic ALTER TABLE. REC_INFO_DEFAULT_ROW_ADD: Renamed from REC_INFO_DEFAULT_ROW. This is for the old instant ADD COLUMN (MDEV-11369) only. REC_INFO_DEFAULT_ROW_ALTER: The more generic metadata record, with additional information for dropped or reordered columns. rec_info_bits_valid(): Remove. The only case when this would fail is when the record is the generic ALTER TABLE metadata record. rec_is_alter_metadata(): Check if a record is the metadata record for instant ALTER TABLE (other than ADD COLUMN). NOTE: This function must not be invoked on node pointer records, because the delete-mark flag in those records may be set (it is garbage), and then a debug assertion could fail because index->is_instant() does not necessarily hold. rec_is_add_metadata(): Check if a record is MDEV-11369 ADD COLUMN metadata record (not more generic instant ALTER TABLE). rec_get_converted_size_comp_prefix_low(): Assume that the metadata field will be stored externally. In dtuple_convert_big_rec() during the rec_get_converted_size() call, it would not be there yet. rec_get_converted_size_comp(): Replace status,fields,n_fields with tuple. rec_init_offsets_comp_ordinary(), rec_get_converted_size_comp_prefix_low(), rec_convert_dtuple_to_rec_comp(): Add template<bool mblob = false>. With mblob=true, process a record with a metadata BLOB. rec_copy_prefix_to_buf(): Assert that no fields beyond the key and system columns are being copied. Exclude the metadata BLOB field. rec_convert_dtuple_to_metadata_comp(): Convert an alter metadata tuple into a record. row_upd_index_replace_metadata(): Apply an update vector to an alter_metadata tuple. row_log_allocate(): Replace dict_index_t::is_instant() with a more appropriate condition that ignores dict_table_t::instant. Only a table on which the MDEV-11369 ADD COLUMN was performed can "lose its instantness" when it becomes empty. After instant DROP COLUMN or reordering columns, we cannot simply convert the table to the canonical format, because the data dictionary cache and all possibly existing references to it from other client connection threads would have to be adjusted. row_quiesce_write_index_fields(): Do not crash when the table contains an instantly dropped column. Thanks to Thirunarayanan Balathandayuthapani for discussing the design and implementing an initial prototype of this. Thanks to Matthias Leich for testing.
7 years ago
MDEV-21907: InnoDB: Enable -Wconversion on clang and GCC The -Wconversion in GCC seems to be stricter than in clang. GCC at least since version 4.4.7 issues truncation warnings for assignments to bitfields, while clang 10 appears to only issue warnings when the sizes in bytes rounded to the nearest integer powers of 2 are different. Before GCC 10.0.0, -Wconversion required more casts and would not allow some operations, such as x<<=1 or x+=1 on a data type that is narrower than int. GCC 5 (but not GCC 4, GCC 6, or any later version) is complaining about x|=y even when x and y are compatible types that are narrower than int. Hence, we must rewrite some x|=y as x=static_cast<byte>(x|y) or similar, or we must disable -Wconversion. In GCC 6 and later, the warning for assigning wider to bitfields that are narrower than 8, 16, or 32 bits can be suppressed by applying a bitwise & with the exact bitmask of the bitfield. For older GCC, we must disable -Wconversion for GCC 4 or 5 in such cases. The bitwise negation operator appears to promote short integers to a wider type, and hence we must add explicit truncation casts around them. Microsoft Visual C does not allow a static_cast to truncate a constant, such as static_cast<byte>(1) truncating int. Hence, we will use the constructor-style cast byte(~1) for such cases. This has been tested at least with GCC 4.8.5, 5.4.0, 7.4.0, 9.2.1, 10.0.0, clang 9.0.1, 10.0.0, and MSVC 14.22.27905 (Microsoft Visual Studio 2019) on 64-bit and 32-bit targets (IA-32, AMD64, POWER 8, POWER 9, ARMv8).
6 years ago
MDEV-15662 Instant DROP COLUMN or changing the order of columns Allow ADD COLUMN anywhere in a table, not only adding as the last column. Allow instant DROP COLUMN and instant changing the order of columns. The added columns will always be added last in clustered index records. In new records, instantly dropped columns will be stored as NULL or empty when possible. Information about dropped and reordered columns will be written in a metadata BLOB (mblob), which is stored before the first 'user' field in the hidden metadata record at the start of the clustered index. The presence of mblob is indicated by setting the delete-mark flag in the metadata record. The metadata BLOB stores the number of clustered index fields, followed by an array of column information for each field. For dropped columns, we store the NOT NULL flag, the fixed length, and for variable-length columns, whether the maximum length exceeded 255 bytes. For non-dropped columns, we store the column position. Unlike with MDEV-11369, when a table becomes empty, it cannot be converted back to the canonical format. The reason for this is that other threads may hold cached objects such as row_prebuilt_t::ins_node that could refer to dropped or reordered index fields. For instant DROP COLUMN and ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC, we must store the n_core_null_bytes in the root page, so that the chain of node pointer records can be followed in order to reach the leftmost leaf page where the metadata record is located. If the mblob is present, we will zero-initialize the strings "infimum" and "supremum" in the root page, and use the last byte of "supremum" for storing the number of null bytes (which are allocated but useless on node pointer pages). This is necessary for btr_cur_instant_init_metadata() to be able to navigate to the mblob. If the PRIMARY KEY contains any variable-length column and some nullable columns were instantly dropped, the dict_index_t::n_nullable in the data dictionary could be smaller than it actually is in the non-leaf pages. Because of this, the non-leaf pages could use more bytes for the null flags than the data dictionary expects, and we could be reading the lengths of the variable-length columns from the wrong offset, and thus reading the child page number from wrong place. This is the result of two design mistakes that involve unnecessary storage of data: First, it is nonsense to store any data fields for the leftmost node pointer records, because the comparisons would be resolved by the MIN_REC_FLAG alone. Second, there cannot be any null fields in the clustered index node pointer fields, but we nevertheless reserve space for all the null flags. Limitations (future work): MDEV-17459 Allow instant ALTER TABLE even if FULLTEXT INDEX exists MDEV-17468 Avoid table rebuild on operations on generated columns MDEV-17494 Refuse ALGORITHM=INSTANT when the row size is too large btr_page_reorganize_low(): Preserve any metadata in the root page. Call lock_move_reorganize_page() only after restoring the "infimum" and "supremum" records, to avoid a memcmp() assertion failure. dict_col_t::DROPPED: Magic value for dict_col_t::ind. dict_col_t::clear_instant(): Renamed from dict_col_t::remove_instant(). Do not assert that the column was instantly added, because we sometimes call this unconditionally for all columns. Convert an instantly added column to a "core column". The old name remove_instant() could be mistaken to refer to "instant DROP COLUMN". dict_col_t::is_added(): Rename from dict_col_t::is_instant(). dtype_t::metadata_blob_init(): Initialize the mblob data type. dtuple_t::is_metadata(), dtuple_t::is_alter_metadata(), upd_t::is_metadata(), upd_t::is_alter_metadata(): Check if info_bits refer to a metadata record. dict_table_t::instant: Metadata about dropped or reordered columns. dict_table_t::prepare_instant(): Prepare ha_innobase_inplace_ctx::instant_table for instant ALTER TABLE. innobase_instant_try() will pass this to dict_table_t::instant_column(). On rollback, dict_table_t::rollback_instant() will be called. dict_table_t::instant_column(): Renamed from instant_add_column(). Add the parameter col_map so that columns can be reordered. Copy and adjust v_cols[] as well. dict_table_t::find(): Find an old column based on a new column number. dict_table_t::serialise_columns(), dict_table_t::deserialise_columns(): Convert the mblob. dict_index_t::instant_metadata(): Create the metadata record for instant ALTER TABLE. Invoke dict_table_t::serialise_columns(). dict_index_t::reconstruct_fields(): Invoked by dict_table_t::deserialise_columns(). dict_index_t::clear_instant_alter(): Move the fields for the dropped columns to the end, and sort the surviving index fields in ascending order of column position. ha_innobase::check_if_supported_inplace_alter(): Do not allow adding a FTS_DOC_ID column if a hidden FTS_DOC_ID column exists due to FULLTEXT INDEX. (This always required ALGORITHM=COPY.) instant_alter_column_possible(): Add a parameter for InnoDB table, to check for additional conditions, such as the maximum number of index fields. ha_innobase_inplace_ctx::first_alter_pos: The first column whose position is affected by instant ADD, DROP, or changing the order of columns. innobase_build_col_map(): Skip added virtual columns. prepare_inplace_add_virtual(): Correctly compute num_to_add_vcol. Remove some unnecessary code. Note that the call to innodb_base_col_setup() should be executed later. commit_try_norebuild(): If ctx->is_instant(), let the virtual columns be added or dropped by innobase_instant_try(). innobase_instant_try(): Fill in a zero default value for the hidden column FTS_DOC_ID (to reduce the work needed in MDEV-17459). If any columns were dropped or reordered (or added not last), delete any SYS_COLUMNS records for the following columns, and insert SYS_COLUMNS records for all subsequent stored columns as well as for all virtual columns. If any virtual column is dropped, rewrite all virtual column metadata. Use a shortcut only for adding virtual columns. This is because innobase_drop_virtual_try() assumes that the dropped virtual columns still exist in ctx->old_table. innodb_update_cols(): Renamed from innodb_update_n_cols(). innobase_add_one_virtual(), innobase_insert_sys_virtual(): Change the return type to bool, and invoke my_error() when detecting an error. innodb_insert_sys_columns(): Insert a record into SYS_COLUMNS. Refactored from innobase_add_one_virtual() and innobase_instant_add_col(). innobase_instant_add_col(): Replace the parameter dfield with type. innobase_instant_drop_cols(): Drop matching columns from SYS_COLUMNS and all columns from SYS_VIRTUAL. innobase_add_virtual_try(), innobase_drop_virtual_try(): Let the caller invoke innodb_update_cols(). innobase_rename_column_try(): Skip dropped columns. commit_cache_norebuild(): Update table->fts->doc_col. dict_mem_table_col_rename_low(): Skip dropped columns. trx_undo_rec_get_partial_row(): Skip dropped columns. trx_undo_update_rec_get_update(): Handle the metadata BLOB correctly. trx_undo_page_report_modify(): Avoid out-of-bounds access to record fields. Log metadata records consistently. Apparently, the first fields of a clustered index may be updated in an update_undo vector when the index is ID_IND of SYS_FOREIGN, as part of renaming the table during ALTER TABLE. Normally, updates of the PRIMARY KEY should be logged as delete-mark and an insert. row_undo_mod_parse_undo_rec(), row_purge_parse_undo_rec(): Use trx_undo_metadata. row_undo_mod_clust_low(): On metadata rollback, roll back the root page too. row_undo_mod_clust(): Relax an assertion. The delete-mark flag was repurposed for ALTER TABLE metadata records. row_rec_to_index_entry_impl(): Add the template parameter mblob and the optional parameter info_bits for specifying the desired new info bits. For the metadata tuple, allow conversion between the original format (ADD COLUMN only) and the generic format (with hidden BLOB). Add the optional parameter "pad" to determine whether the tuple should be padded to the index fields (on ALTER TABLE it should), or whether it should remain at its original size (on rollback). row_build_index_entry_low(): Clean up the code, removing redundant variables and conditions. For instantly dropped columns, generate a dummy value that is NULL, the empty string, or a fixed length of NUL bytes, depending on the type of the dropped column. row_upd_clust_rec_by_insert_inherit_func(): On the update of PRIMARY KEY of a record that contained a dropped column whose value was stored externally, we will be inserting a dummy NULL or empty string value to the field of the dropped column. The externally stored column would eventually be dropped when purge removes the delete-marked record for the old PRIMARY KEY value. btr_index_rec_validate(): Recognize the metadata record. btr_discard_only_page_on_level(): Preserve the generic instant ALTER TABLE metadata. btr_set_instant(): Replaces page_set_instant(). This sets a clustered index root page to the appropriate format, or upgrades from the MDEV-11369 instant ADD COLUMN to generic ALTER TABLE format. btr_cur_instant_init_low(): Read and validate the metadata BLOB page before reconstructing the dictionary information based on it. btr_cur_instant_init_metadata(): Do not read any lengths from the metadata record header before reading the BLOB. At this point, we would not actually know how many nullable fields the metadata record contains. btr_cur_instant_root_init(): Initialize n_core_null_bytes in one of two possible ways. btr_cur_trim(): Handle the mblob record. row_metadata_to_tuple(): Convert a metadata record to a data tuple, based on the new info_bits of the metadata record. btr_cur_pessimistic_update(): Invoke row_metadata_to_tuple() if needed. Invoke dtuple_convert_big_rec() for metadata records if the record is too large, or if the mblob is not yet marked as externally stored. btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete(): When the last user record is deleted, do not delete the generic instant ALTER TABLE metadata record. Only delete MDEV-11369 instant ADD COLUMN metadata records. btr_cur_optimistic_insert(): Avoid unnecessary computation of rec_size. btr_pcur_store_position(): Allow a logically empty page to contain a metadata record for generic ALTER TABLE. REC_INFO_DEFAULT_ROW_ADD: Renamed from REC_INFO_DEFAULT_ROW. This is for the old instant ADD COLUMN (MDEV-11369) only. REC_INFO_DEFAULT_ROW_ALTER: The more generic metadata record, with additional information for dropped or reordered columns. rec_info_bits_valid(): Remove. The only case when this would fail is when the record is the generic ALTER TABLE metadata record. rec_is_alter_metadata(): Check if a record is the metadata record for instant ALTER TABLE (other than ADD COLUMN). NOTE: This function must not be invoked on node pointer records, because the delete-mark flag in those records may be set (it is garbage), and then a debug assertion could fail because index->is_instant() does not necessarily hold. rec_is_add_metadata(): Check if a record is MDEV-11369 ADD COLUMN metadata record (not more generic instant ALTER TABLE). rec_get_converted_size_comp_prefix_low(): Assume that the metadata field will be stored externally. In dtuple_convert_big_rec() during the rec_get_converted_size() call, it would not be there yet. rec_get_converted_size_comp(): Replace status,fields,n_fields with tuple. rec_init_offsets_comp_ordinary(), rec_get_converted_size_comp_prefix_low(), rec_convert_dtuple_to_rec_comp(): Add template<bool mblob = false>. With mblob=true, process a record with a metadata BLOB. rec_copy_prefix_to_buf(): Assert that no fields beyond the key and system columns are being copied. Exclude the metadata BLOB field. rec_convert_dtuple_to_metadata_comp(): Convert an alter metadata tuple into a record. row_upd_index_replace_metadata(): Apply an update vector to an alter_metadata tuple. row_log_allocate(): Replace dict_index_t::is_instant() with a more appropriate condition that ignores dict_table_t::instant. Only a table on which the MDEV-11369 ADD COLUMN was performed can "lose its instantness" when it becomes empty. After instant DROP COLUMN or reordering columns, we cannot simply convert the table to the canonical format, because the data dictionary cache and all possibly existing references to it from other client connection threads would have to be adjusted. row_quiesce_write_index_fields(): Do not crash when the table contains an instantly dropped column. Thanks to Thirunarayanan Balathandayuthapani for discussing the design and implementing an initial prototype of this. Thanks to Matthias Leich for testing.
7 years ago
MDEV-15662 Instant DROP COLUMN or changing the order of columns Allow ADD COLUMN anywhere in a table, not only adding as the last column. Allow instant DROP COLUMN and instant changing the order of columns. The added columns will always be added last in clustered index records. In new records, instantly dropped columns will be stored as NULL or empty when possible. Information about dropped and reordered columns will be written in a metadata BLOB (mblob), which is stored before the first 'user' field in the hidden metadata record at the start of the clustered index. The presence of mblob is indicated by setting the delete-mark flag in the metadata record. The metadata BLOB stores the number of clustered index fields, followed by an array of column information for each field. For dropped columns, we store the NOT NULL flag, the fixed length, and for variable-length columns, whether the maximum length exceeded 255 bytes. For non-dropped columns, we store the column position. Unlike with MDEV-11369, when a table becomes empty, it cannot be converted back to the canonical format. The reason for this is that other threads may hold cached objects such as row_prebuilt_t::ins_node that could refer to dropped or reordered index fields. For instant DROP COLUMN and ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC, we must store the n_core_null_bytes in the root page, so that the chain of node pointer records can be followed in order to reach the leftmost leaf page where the metadata record is located. If the mblob is present, we will zero-initialize the strings "infimum" and "supremum" in the root page, and use the last byte of "supremum" for storing the number of null bytes (which are allocated but useless on node pointer pages). This is necessary for btr_cur_instant_init_metadata() to be able to navigate to the mblob. If the PRIMARY KEY contains any variable-length column and some nullable columns were instantly dropped, the dict_index_t::n_nullable in the data dictionary could be smaller than it actually is in the non-leaf pages. Because of this, the non-leaf pages could use more bytes for the null flags than the data dictionary expects, and we could be reading the lengths of the variable-length columns from the wrong offset, and thus reading the child page number from wrong place. This is the result of two design mistakes that involve unnecessary storage of data: First, it is nonsense to store any data fields for the leftmost node pointer records, because the comparisons would be resolved by the MIN_REC_FLAG alone. Second, there cannot be any null fields in the clustered index node pointer fields, but we nevertheless reserve space for all the null flags. Limitations (future work): MDEV-17459 Allow instant ALTER TABLE even if FULLTEXT INDEX exists MDEV-17468 Avoid table rebuild on operations on generated columns MDEV-17494 Refuse ALGORITHM=INSTANT when the row size is too large btr_page_reorganize_low(): Preserve any metadata in the root page. Call lock_move_reorganize_page() only after restoring the "infimum" and "supremum" records, to avoid a memcmp() assertion failure. dict_col_t::DROPPED: Magic value for dict_col_t::ind. dict_col_t::clear_instant(): Renamed from dict_col_t::remove_instant(). Do not assert that the column was instantly added, because we sometimes call this unconditionally for all columns. Convert an instantly added column to a "core column". The old name remove_instant() could be mistaken to refer to "instant DROP COLUMN". dict_col_t::is_added(): Rename from dict_col_t::is_instant(). dtype_t::metadata_blob_init(): Initialize the mblob data type. dtuple_t::is_metadata(), dtuple_t::is_alter_metadata(), upd_t::is_metadata(), upd_t::is_alter_metadata(): Check if info_bits refer to a metadata record. dict_table_t::instant: Metadata about dropped or reordered columns. dict_table_t::prepare_instant(): Prepare ha_innobase_inplace_ctx::instant_table for instant ALTER TABLE. innobase_instant_try() will pass this to dict_table_t::instant_column(). On rollback, dict_table_t::rollback_instant() will be called. dict_table_t::instant_column(): Renamed from instant_add_column(). Add the parameter col_map so that columns can be reordered. Copy and adjust v_cols[] as well. dict_table_t::find(): Find an old column based on a new column number. dict_table_t::serialise_columns(), dict_table_t::deserialise_columns(): Convert the mblob. dict_index_t::instant_metadata(): Create the metadata record for instant ALTER TABLE. Invoke dict_table_t::serialise_columns(). dict_index_t::reconstruct_fields(): Invoked by dict_table_t::deserialise_columns(). dict_index_t::clear_instant_alter(): Move the fields for the dropped columns to the end, and sort the surviving index fields in ascending order of column position. ha_innobase::check_if_supported_inplace_alter(): Do not allow adding a FTS_DOC_ID column if a hidden FTS_DOC_ID column exists due to FULLTEXT INDEX. (This always required ALGORITHM=COPY.) instant_alter_column_possible(): Add a parameter for InnoDB table, to check for additional conditions, such as the maximum number of index fields. ha_innobase_inplace_ctx::first_alter_pos: The first column whose position is affected by instant ADD, DROP, or changing the order of columns. innobase_build_col_map(): Skip added virtual columns. prepare_inplace_add_virtual(): Correctly compute num_to_add_vcol. Remove some unnecessary code. Note that the call to innodb_base_col_setup() should be executed later. commit_try_norebuild(): If ctx->is_instant(), let the virtual columns be added or dropped by innobase_instant_try(). innobase_instant_try(): Fill in a zero default value for the hidden column FTS_DOC_ID (to reduce the work needed in MDEV-17459). If any columns were dropped or reordered (or added not last), delete any SYS_COLUMNS records for the following columns, and insert SYS_COLUMNS records for all subsequent stored columns as well as for all virtual columns. If any virtual column is dropped, rewrite all virtual column metadata. Use a shortcut only for adding virtual columns. This is because innobase_drop_virtual_try() assumes that the dropped virtual columns still exist in ctx->old_table. innodb_update_cols(): Renamed from innodb_update_n_cols(). innobase_add_one_virtual(), innobase_insert_sys_virtual(): Change the return type to bool, and invoke my_error() when detecting an error. innodb_insert_sys_columns(): Insert a record into SYS_COLUMNS. Refactored from innobase_add_one_virtual() and innobase_instant_add_col(). innobase_instant_add_col(): Replace the parameter dfield with type. innobase_instant_drop_cols(): Drop matching columns from SYS_COLUMNS and all columns from SYS_VIRTUAL. innobase_add_virtual_try(), innobase_drop_virtual_try(): Let the caller invoke innodb_update_cols(). innobase_rename_column_try(): Skip dropped columns. commit_cache_norebuild(): Update table->fts->doc_col. dict_mem_table_col_rename_low(): Skip dropped columns. trx_undo_rec_get_partial_row(): Skip dropped columns. trx_undo_update_rec_get_update(): Handle the metadata BLOB correctly. trx_undo_page_report_modify(): Avoid out-of-bounds access to record fields. Log metadata records consistently. Apparently, the first fields of a clustered index may be updated in an update_undo vector when the index is ID_IND of SYS_FOREIGN, as part of renaming the table during ALTER TABLE. Normally, updates of the PRIMARY KEY should be logged as delete-mark and an insert. row_undo_mod_parse_undo_rec(), row_purge_parse_undo_rec(): Use trx_undo_metadata. row_undo_mod_clust_low(): On metadata rollback, roll back the root page too. row_undo_mod_clust(): Relax an assertion. The delete-mark flag was repurposed for ALTER TABLE metadata records. row_rec_to_index_entry_impl(): Add the template parameter mblob and the optional parameter info_bits for specifying the desired new info bits. For the metadata tuple, allow conversion between the original format (ADD COLUMN only) and the generic format (with hidden BLOB). Add the optional parameter "pad" to determine whether the tuple should be padded to the index fields (on ALTER TABLE it should), or whether it should remain at its original size (on rollback). row_build_index_entry_low(): Clean up the code, removing redundant variables and conditions. For instantly dropped columns, generate a dummy value that is NULL, the empty string, or a fixed length of NUL bytes, depending on the type of the dropped column. row_upd_clust_rec_by_insert_inherit_func(): On the update of PRIMARY KEY of a record that contained a dropped column whose value was stored externally, we will be inserting a dummy NULL or empty string value to the field of the dropped column. The externally stored column would eventually be dropped when purge removes the delete-marked record for the old PRIMARY KEY value. btr_index_rec_validate(): Recognize the metadata record. btr_discard_only_page_on_level(): Preserve the generic instant ALTER TABLE metadata. btr_set_instant(): Replaces page_set_instant(). This sets a clustered index root page to the appropriate format, or upgrades from the MDEV-11369 instant ADD COLUMN to generic ALTER TABLE format. btr_cur_instant_init_low(): Read and validate the metadata BLOB page before reconstructing the dictionary information based on it. btr_cur_instant_init_metadata(): Do not read any lengths from the metadata record header before reading the BLOB. At this point, we would not actually know how many nullable fields the metadata record contains. btr_cur_instant_root_init(): Initialize n_core_null_bytes in one of two possible ways. btr_cur_trim(): Handle the mblob record. row_metadata_to_tuple(): Convert a metadata record to a data tuple, based on the new info_bits of the metadata record. btr_cur_pessimistic_update(): Invoke row_metadata_to_tuple() if needed. Invoke dtuple_convert_big_rec() for metadata records if the record is too large, or if the mblob is not yet marked as externally stored. btr_cur_optimistic_delete_func(), btr_cur_pessimistic_delete(): When the last user record is deleted, do not delete the generic instant ALTER TABLE metadata record. Only delete MDEV-11369 instant ADD COLUMN metadata records. btr_cur_optimistic_insert(): Avoid unnecessary computation of rec_size. btr_pcur_store_position(): Allow a logically empty page to contain a metadata record for generic ALTER TABLE. REC_INFO_DEFAULT_ROW_ADD: Renamed from REC_INFO_DEFAULT_ROW. This is for the old instant ADD COLUMN (MDEV-11369) only. REC_INFO_DEFAULT_ROW_ALTER: The more generic metadata record, with additional information for dropped or reordered columns. rec_info_bits_valid(): Remove. The only case when this would fail is when the record is the generic ALTER TABLE metadata record. rec_is_alter_metadata(): Check if a record is the metadata record for instant ALTER TABLE (other than ADD COLUMN). NOTE: This function must not be invoked on node pointer records, because the delete-mark flag in those records may be set (it is garbage), and then a debug assertion could fail because index->is_instant() does not necessarily hold. rec_is_add_metadata(): Check if a record is MDEV-11369 ADD COLUMN metadata record (not more generic instant ALTER TABLE). rec_get_converted_size_comp_prefix_low(): Assume that the metadata field will be stored externally. In dtuple_convert_big_rec() during the rec_get_converted_size() call, it would not be there yet. rec_get_converted_size_comp(): Replace status,fields,n_fields with tuple. rec_init_offsets_comp_ordinary(), rec_get_converted_size_comp_prefix_low(), rec_convert_dtuple_to_rec_comp(): Add template<bool mblob = false>. With mblob=true, process a record with a metadata BLOB. rec_copy_prefix_to_buf(): Assert that no fields beyond the key and system columns are being copied. Exclude the metadata BLOB field. rec_convert_dtuple_to_metadata_comp(): Convert an alter metadata tuple into a record. row_upd_index_replace_metadata(): Apply an update vector to an alter_metadata tuple. row_log_allocate(): Replace dict_index_t::is_instant() with a more appropriate condition that ignores dict_table_t::instant. Only a table on which the MDEV-11369 ADD COLUMN was performed can "lose its instantness" when it becomes empty. After instant DROP COLUMN or reordering columns, we cannot simply convert the table to the canonical format, because the data dictionary cache and all possibly existing references to it from other client connection threads would have to be adjusted. row_quiesce_write_index_fields(): Do not crash when the table contains an instantly dropped column. Thanks to Thirunarayanan Balathandayuthapani for discussing the design and implementing an initial prototype of this. Thanks to Matthias Leich for testing.
7 years ago
11 years ago
  1. /*****************************************************************************
  2. Copyright (c) 1996, 2017, Oracle and/or its affiliates. All Rights Reserved.
  3. Copyright (c) 2012, Facebook Inc.
  4. Copyright (c) 2013, 2021, MariaDB Corporation.
  5. This program is free software; you can redistribute it and/or modify it under
  6. the terms of the GNU General Public License as published by the Free Software
  7. Foundation; version 2 of the License.
  8. This program is distributed in the hope that it will be useful, but WITHOUT
  9. ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
  10. FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
  11. You should have received a copy of the GNU General Public License along with
  12. this program; if not, write to the Free Software Foundation, Inc.,
  13. 51 Franklin Street, Fifth Floor, Boston, MA 02110-1335 USA
  14. *****************************************************************************/
  15. /**************************************************//**
  16. @file include/dict0mem.h
  17. Data dictionary memory object creation
  18. Created 1/8/1996 Heikki Tuuri
  19. *******************************************************/
  20. #ifndef dict0mem_h
  21. #define dict0mem_h
  22. #include "data0type.h"
  23. #include "mem0mem.h"
  24. #include "row0types.h"
  25. #include "rem0types.h"
  26. #include "btr0types.h"
  27. #include "lock0types.h"
  28. #include "que0types.h"
  29. #include "sync0rw.h"
  30. #include "ut0mem.h"
  31. #include "ut0rnd.h"
  32. #include "ut0byte.h"
  33. #include "hash0hash.h"
  34. #include "trx0types.h"
  35. #include "fts0fts.h"
  36. #include "buf0buf.h"
  37. #include "gis0type.h"
  38. #include "fil0fil.h"
  39. #include "fil0crypt.h"
  40. #include "mysql_com.h"
  41. #include <sql_const.h>
  42. #include <set>
  43. #include <algorithm>
  44. #include <iterator>
  45. #include <ostream>
  46. #include <mutex>
  47. /* Forward declaration. */
  48. struct ib_rbt_t;
  49. /** Type flags of an index: OR'ing of the flags is allowed to define a
  50. combination of types */
  51. /* @{ */
  52. #define DICT_CLUSTERED 1 /*!< clustered index; for other than
  53. auto-generated clustered indexes,
  54. also DICT_UNIQUE will be set */
  55. #define DICT_UNIQUE 2 /*!< unique index */
  56. #define DICT_IBUF 8 /*!< insert buffer tree */
  57. #define DICT_CORRUPT 16 /*!< bit to store the corrupted flag
  58. in SYS_INDEXES.TYPE */
  59. #define DICT_FTS 32 /* FTS index; can't be combined with the
  60. other flags */
  61. #define DICT_SPATIAL 64 /* SPATIAL index; can't be combined with the
  62. other flags */
  63. #define DICT_VIRTUAL 128 /* Index on Virtual column */
  64. #define DICT_IT_BITS 8 /*!< number of bits used for
  65. SYS_INDEXES.TYPE */
  66. /* @} */
  67. #if 0 /* not implemented, retained for history */
  68. /** Types for a table object */
  69. #define DICT_TABLE_ORDINARY 1 /*!< ordinary table */
  70. #define DICT_TABLE_CLUSTER_MEMBER 2
  71. #define DICT_TABLE_CLUSTER 3 /* this means that the table is
  72. really a cluster definition */
  73. #endif
  74. /* Table and tablespace flags are generally not used for the Antelope file
  75. format except for the low order bit, which is used differently depending on
  76. where the flags are stored.
  77. ==================== Low order flags bit =========================
  78. | REDUNDANT | COMPACT | COMPRESSED and DYNAMIC
  79. SYS_TABLES.TYPE | 1 | 1 | 1
  80. dict_table_t::flags | 0 | 1 | 1
  81. FSP_SPACE_FLAGS | 0 | 0 | 1
  82. fil_space_t::flags | 0 | 0 | 1
  83. Before the 5.1 plugin, SYS_TABLES.TYPE was always DICT_TABLE_ORDINARY (1)
  84. and the tablespace flags field was always 0. In the 5.1 plugin, these fields
  85. were repurposed to identify compressed and dynamic row formats.
  86. The following types and constants describe the flags found in dict_table_t
  87. and SYS_TABLES.TYPE. Similar flags found in fil_space_t and FSP_SPACE_FLAGS
  88. are described in fsp0fsp.h. */
  89. /* @{ */
  90. /** dict_table_t::flags bit 0 is equal to 0 if the row format = Redundant */
  91. #define DICT_TF_REDUNDANT 0 /*!< Redundant row format. */
  92. /** dict_table_t::flags bit 0 is equal to 1 if the row format = Compact */
  93. #define DICT_TF_COMPACT 1U /*!< Compact row format. */
  94. /** This bitmask is used in SYS_TABLES.N_COLS to set and test whether
  95. the Compact page format is used, i.e ROW_FORMAT != REDUNDANT */
  96. #define DICT_N_COLS_COMPACT 0x80000000UL
  97. /** Width of the COMPACT flag */
  98. #define DICT_TF_WIDTH_COMPACT 1
  99. /** Width of the ZIP_SSIZE flag */
  100. #define DICT_TF_WIDTH_ZIP_SSIZE 4
  101. /** Width of the ATOMIC_BLOBS flag. The ROW_FORMAT=REDUNDANT and
  102. ROW_FORMAT=COMPACT broke up BLOB and TEXT fields, storing the first 768 bytes
  103. in the clustered index. ROW_FORMAT=DYNAMIC and ROW_FORMAT=COMPRESSED
  104. store the whole blob or text field off-page atomically.
  105. Secondary indexes are created from this external data using row_ext_t
  106. to cache the BLOB prefixes. */
  107. #define DICT_TF_WIDTH_ATOMIC_BLOBS 1
  108. /** If a table is created with the MYSQL option DATA DIRECTORY and
  109. innodb-file-per-table, an older engine will not be able to find that table.
  110. This flag prevents older engines from attempting to open the table and
  111. allows InnoDB to update_create_info() accordingly. */
  112. #define DICT_TF_WIDTH_DATA_DIR 1
  113. /**
  114. Width of the page compression flag
  115. */
  116. #define DICT_TF_WIDTH_PAGE_COMPRESSION 1
  117. #define DICT_TF_WIDTH_PAGE_COMPRESSION_LEVEL 4
  118. /**
  119. The NO_ROLLBACK flag (3=yes; the values 1,2 used stand for
  120. ATOMIC_WRITES=ON and ATOMIC_WRITES=OFF between MariaDB 10.1.0 and 10.2.3)
  121. */
  122. #define DICT_TF_WIDTH_NO_ROLLBACK 2
  123. /** Width of all the currently known table flags */
  124. #define DICT_TF_BITS (DICT_TF_WIDTH_COMPACT \
  125. + DICT_TF_WIDTH_ZIP_SSIZE \
  126. + DICT_TF_WIDTH_ATOMIC_BLOBS \
  127. + DICT_TF_WIDTH_DATA_DIR \
  128. + DICT_TF_WIDTH_PAGE_COMPRESSION \
  129. + DICT_TF_WIDTH_PAGE_COMPRESSION_LEVEL \
  130. + DICT_TF_WIDTH_NO_ROLLBACK)
  131. /** Zero relative shift position of the COMPACT field */
  132. #define DICT_TF_POS_COMPACT 0
  133. /** Zero relative shift position of the ZIP_SSIZE field */
  134. #define DICT_TF_POS_ZIP_SSIZE (DICT_TF_POS_COMPACT \
  135. + DICT_TF_WIDTH_COMPACT)
  136. /** Zero relative shift position of the ATOMIC_BLOBS field */
  137. #define DICT_TF_POS_ATOMIC_BLOBS (DICT_TF_POS_ZIP_SSIZE \
  138. + DICT_TF_WIDTH_ZIP_SSIZE)
  139. /** Zero relative shift position of the DATA_DIR field */
  140. #define DICT_TF_POS_DATA_DIR (DICT_TF_POS_ATOMIC_BLOBS \
  141. + DICT_TF_WIDTH_ATOMIC_BLOBS)
  142. /** Zero relative shift position of the PAGE_COMPRESSION field */
  143. #define DICT_TF_POS_PAGE_COMPRESSION (DICT_TF_POS_DATA_DIR \
  144. + DICT_TF_WIDTH_DATA_DIR)
  145. /** Zero relative shift position of the PAGE_COMPRESSION_LEVEL field */
  146. #define DICT_TF_POS_PAGE_COMPRESSION_LEVEL (DICT_TF_POS_PAGE_COMPRESSION \
  147. + DICT_TF_WIDTH_PAGE_COMPRESSION)
  148. /** Zero relative shift position of the NO_ROLLBACK field */
  149. #define DICT_TF_POS_NO_ROLLBACK (DICT_TF_POS_PAGE_COMPRESSION_LEVEL \
  150. + DICT_TF_WIDTH_PAGE_COMPRESSION_LEVEL)
  151. #define DICT_TF_POS_UNUSED (DICT_TF_POS_NO_ROLLBACK \
  152. + DICT_TF_WIDTH_NO_ROLLBACK)
  153. /** Bit mask of the COMPACT field */
  154. #define DICT_TF_MASK_COMPACT \
  155. ((~(~0U << DICT_TF_WIDTH_COMPACT)) \
  156. << DICT_TF_POS_COMPACT)
  157. /** Bit mask of the ZIP_SSIZE field */
  158. #define DICT_TF_MASK_ZIP_SSIZE \
  159. ((~(~0U << DICT_TF_WIDTH_ZIP_SSIZE)) \
  160. << DICT_TF_POS_ZIP_SSIZE)
  161. /** Bit mask of the ATOMIC_BLOBS field */
  162. #define DICT_TF_MASK_ATOMIC_BLOBS \
  163. ((~(~0U << DICT_TF_WIDTH_ATOMIC_BLOBS)) \
  164. << DICT_TF_POS_ATOMIC_BLOBS)
  165. /** Bit mask of the DATA_DIR field */
  166. #define DICT_TF_MASK_DATA_DIR \
  167. ((~(~0U << DICT_TF_WIDTH_DATA_DIR)) \
  168. << DICT_TF_POS_DATA_DIR)
  169. /** Bit mask of the PAGE_COMPRESSION field */
  170. #define DICT_TF_MASK_PAGE_COMPRESSION \
  171. ((~(~0U << DICT_TF_WIDTH_PAGE_COMPRESSION)) \
  172. << DICT_TF_POS_PAGE_COMPRESSION)
  173. /** Bit mask of the PAGE_COMPRESSION_LEVEL field */
  174. #define DICT_TF_MASK_PAGE_COMPRESSION_LEVEL \
  175. ((~(~0U << DICT_TF_WIDTH_PAGE_COMPRESSION_LEVEL)) \
  176. << DICT_TF_POS_PAGE_COMPRESSION_LEVEL)
  177. /** Bit mask of the NO_ROLLBACK field */
  178. #define DICT_TF_MASK_NO_ROLLBACK \
  179. ((~(~0U << DICT_TF_WIDTH_NO_ROLLBACK)) \
  180. << DICT_TF_POS_NO_ROLLBACK)
  181. /** Return the value of the COMPACT field */
  182. #define DICT_TF_GET_COMPACT(flags) \
  183. ((flags & DICT_TF_MASK_COMPACT) \
  184. >> DICT_TF_POS_COMPACT)
  185. /** Return the value of the ZIP_SSIZE field */
  186. #define DICT_TF_GET_ZIP_SSIZE(flags) \
  187. ((flags & DICT_TF_MASK_ZIP_SSIZE) \
  188. >> DICT_TF_POS_ZIP_SSIZE)
  189. /** Return the value of the ATOMIC_BLOBS field */
  190. #define DICT_TF_HAS_ATOMIC_BLOBS(flags) \
  191. ((flags & DICT_TF_MASK_ATOMIC_BLOBS) \
  192. >> DICT_TF_POS_ATOMIC_BLOBS)
  193. /** Return the value of the DATA_DIR field */
  194. #define DICT_TF_HAS_DATA_DIR(flags) \
  195. ((flags & DICT_TF_MASK_DATA_DIR) \
  196. >> DICT_TF_POS_DATA_DIR)
  197. /** Return the value of the PAGE_COMPRESSION field */
  198. #define DICT_TF_GET_PAGE_COMPRESSION(flags) \
  199. ((flags & DICT_TF_MASK_PAGE_COMPRESSION) \
  200. >> DICT_TF_POS_PAGE_COMPRESSION)
  201. /** Return the value of the PAGE_COMPRESSION_LEVEL field */
  202. #define DICT_TF_GET_PAGE_COMPRESSION_LEVEL(flags) \
  203. ((flags & DICT_TF_MASK_PAGE_COMPRESSION_LEVEL) \
  204. >> DICT_TF_POS_PAGE_COMPRESSION_LEVEL)
  205. /* @} */
  206. /** @brief Table Flags set number 2.
  207. These flags will be stored in SYS_TABLES.MIX_LEN. All unused flags
  208. will be written as 0. The column may contain garbage for tables
  209. created with old versions of InnoDB that only implemented
  210. ROW_FORMAT=REDUNDANT. InnoDB engines do not check these flags
  211. for unknown bits in order to protect backward incompatibility. */
  212. /* @{ */
  213. /** Total number of bits in table->flags2. */
  214. #define DICT_TF2_BITS 7
  215. #define DICT_TF2_UNUSED_BIT_MASK (~0U << DICT_TF2_BITS)
  216. #define DICT_TF2_BIT_MASK ~DICT_TF2_UNUSED_BIT_MASK
  217. /** TEMPORARY; TRUE for tables from CREATE TEMPORARY TABLE. */
  218. #define DICT_TF2_TEMPORARY 1U
  219. /** The table has an internal defined DOC ID column */
  220. #define DICT_TF2_FTS_HAS_DOC_ID 2U
  221. /** The table has an FTS index */
  222. #define DICT_TF2_FTS 4U
  223. /** Need to add Doc ID column for FTS index build.
  224. This is a transient bit for index build */
  225. #define DICT_TF2_FTS_ADD_DOC_ID 8U
  226. /** This bit is used during table creation to indicate that it will
  227. use its own tablespace instead of the system tablespace. */
  228. #define DICT_TF2_USE_FILE_PER_TABLE 16U
  229. /** Set when we discard/detach the tablespace */
  230. #define DICT_TF2_DISCARDED 32U
  231. /** This bit is set if all aux table names (both common tables and
  232. index tables) of a FTS table are in HEX format. */
  233. #define DICT_TF2_FTS_AUX_HEX_NAME 64U
  234. /* @} */
  235. #define DICT_TF2_FLAG_SET(table, flag) \
  236. (table->flags2 |= (flag))
  237. #define DICT_TF2_FLAG_IS_SET(table, flag) \
  238. (table->flags2 & (flag))
  239. #define DICT_TF2_FLAG_UNSET(table, flag) \
  240. (table->flags2 &= ~(flag) & ((1U << DICT_TF2_BITS) - 1))
  241. /** Tables could be chained together with Foreign key constraint. When
  242. first load the parent table, we would load all of its descedents.
  243. This could result in rescursive calls and out of stack error eventually.
  244. DICT_FK_MAX_RECURSIVE_LOAD defines the maximum number of recursive loads,
  245. when exceeded, the child table will not be loaded. It will be loaded when
  246. the foreign constraint check needs to be run. */
  247. #define DICT_FK_MAX_RECURSIVE_LOAD 20
  248. /** Similarly, when tables are chained together with foreign key constraints
  249. with on cascading delete/update clause, delete from parent table could
  250. result in recursive cascading calls. This defines the maximum number of
  251. such cascading deletes/updates allowed. When exceeded, the delete from
  252. parent table will fail, and user has to drop excessive foreign constraint
  253. before proceeds. */
  254. #define FK_MAX_CASCADE_DEL 15
  255. /** Create a table memory object.
  256. @param name table name
  257. @param space tablespace
  258. @param n_cols total number of columns (both virtual and non-virtual)
  259. @param n_v_cols number of virtual columns
  260. @param flags table flags
  261. @param flags2 table flags2
  262. @return own: table object */
  263. dict_table_t *dict_mem_table_create(const char *name, fil_space_t *space,
  264. ulint n_cols, ulint n_v_cols, ulint flags,
  265. ulint flags2);
  266. /****************************************************************/ /**
  267. Free a table memory object. */
  268. void
  269. dict_mem_table_free(
  270. /*================*/
  271. dict_table_t* table); /*!< in: table */
  272. /**********************************************************************//**
  273. Adds a column definition to a table. */
  274. void
  275. dict_mem_table_add_col(
  276. /*===================*/
  277. dict_table_t* table, /*!< in: table */
  278. mem_heap_t* heap, /*!< in: temporary memory heap, or NULL */
  279. const char* name, /*!< in: column name, or NULL */
  280. ulint mtype, /*!< in: main datatype */
  281. ulint prtype, /*!< in: precise type */
  282. ulint len) /*!< in: precision */
  283. MY_ATTRIBUTE((nonnull(1)));
  284. /** Adds a virtual column definition to a table.
  285. @param[in,out] table table
  286. @param[in] heap temporary memory heap, or NULL. It is
  287. used to store name when we have not finished
  288. adding all columns. When all columns are
  289. added, the whole name will copy to memory from
  290. table->heap
  291. @param[in] name column name
  292. @param[in] mtype main datatype
  293. @param[in] prtype precise type
  294. @param[in] len length
  295. @param[in] pos position in a table
  296. @param[in] num_base number of base columns
  297. @return the virtual column definition */
  298. dict_v_col_t*
  299. dict_mem_table_add_v_col(
  300. dict_table_t* table,
  301. mem_heap_t* heap,
  302. const char* name,
  303. ulint mtype,
  304. ulint prtype,
  305. ulint len,
  306. ulint pos,
  307. ulint num_base);
  308. /** Adds a stored column definition to a table.
  309. @param[in] table table
  310. @param[in] num_base number of base columns. */
  311. void
  312. dict_mem_table_add_s_col(
  313. dict_table_t* table,
  314. ulint num_base);
  315. /**********************************************************************//**
  316. Renames a column of a table in the data dictionary cache. */
  317. void
  318. dict_mem_table_col_rename(
  319. /*======================*/
  320. dict_table_t* table, /*!< in/out: table */
  321. ulint nth_col,/*!< in: column index */
  322. const char* from, /*!< in: old column name */
  323. const char* to, /*!< in: new column name */
  324. bool is_virtual);
  325. /*!< in: if this is a virtual column */
  326. /**********************************************************************//**
  327. This function populates a dict_col_t memory structure with
  328. supplied information. */
  329. void
  330. dict_mem_fill_column_struct(
  331. /*========================*/
  332. dict_col_t* column, /*!< out: column struct to be
  333. filled */
  334. ulint col_pos, /*!< in: column position */
  335. ulint mtype, /*!< in: main data type */
  336. ulint prtype, /*!< in: precise type */
  337. ulint col_len); /*!< in: column length */
  338. /**********************************************************************//**
  339. This function poplulates a dict_index_t index memory structure with
  340. supplied information. */
  341. UNIV_INLINE
  342. void
  343. dict_mem_fill_index_struct(
  344. /*=======================*/
  345. dict_index_t* index, /*!< out: index to be filled */
  346. mem_heap_t* heap, /*!< in: memory heap */
  347. const char* index_name, /*!< in: index name */
  348. ulint type, /*!< in: DICT_UNIQUE,
  349. DICT_CLUSTERED, ... ORed */
  350. ulint n_fields); /*!< in: number of fields */
  351. /**********************************************************************//**
  352. Creates an index memory object.
  353. @return own: index object */
  354. dict_index_t*
  355. dict_mem_index_create(
  356. /*==================*/
  357. dict_table_t* table, /*!< in: table */
  358. const char* index_name, /*!< in: index name */
  359. ulint type, /*!< in: DICT_UNIQUE,
  360. DICT_CLUSTERED, ... ORed */
  361. ulint n_fields); /*!< in: number of fields */
  362. /**********************************************************************//**
  363. Adds a field definition to an index. NOTE: does not take a copy
  364. of the column name if the field is a column. The memory occupied
  365. by the column name may be released only after publishing the index. */
  366. void
  367. dict_mem_index_add_field(
  368. /*=====================*/
  369. dict_index_t* index, /*!< in: index */
  370. const char* name, /*!< in: column name */
  371. ulint prefix_len); /*!< in: 0 or the column prefix length
  372. in a MySQL index like
  373. INDEX (textcol(25)) */
  374. /**********************************************************************//**
  375. Frees an index memory object. */
  376. void
  377. dict_mem_index_free(
  378. /*================*/
  379. dict_index_t* index); /*!< in: index */
  380. /**********************************************************************//**
  381. Creates and initializes a foreign constraint memory object.
  382. @return own: foreign constraint struct */
  383. dict_foreign_t*
  384. dict_mem_foreign_create(void);
  385. /*=========================*/
  386. /**********************************************************************//**
  387. Sets the foreign_table_name_lookup pointer based on the value of
  388. lower_case_table_names. If that is 0 or 1, foreign_table_name_lookup
  389. will point to foreign_table_name. If 2, then another string is
  390. allocated from the heap and set to lower case. */
  391. void
  392. dict_mem_foreign_table_name_lookup_set(
  393. /*===================================*/
  394. dict_foreign_t* foreign, /*!< in/out: foreign struct */
  395. ibool do_alloc); /*!< in: is an alloc needed */
  396. /**********************************************************************//**
  397. Sets the referenced_table_name_lookup pointer based on the value of
  398. lower_case_table_names. If that is 0 or 1, referenced_table_name_lookup
  399. will point to referenced_table_name. If 2, then another string is
  400. allocated from the heap and set to lower case. */
  401. void
  402. dict_mem_referenced_table_name_lookup_set(
  403. /*======================================*/
  404. dict_foreign_t* foreign, /*!< in/out: foreign struct */
  405. ibool do_alloc); /*!< in: is an alloc needed */
  406. /** Fills the dependent virtual columns in a set.
  407. Reason for being dependent are
  408. 1) FK can be present on base column of virtual columns
  409. 2) FK can be present on column which is a part of virtual index
  410. @param[in,out] foreign foreign key information. */
  411. void
  412. dict_mem_foreign_fill_vcol_set(
  413. dict_foreign_t* foreign);
  414. /** Fill virtual columns set in each fk constraint present in the table.
  415. @param[in,out] table innodb table object. */
  416. void
  417. dict_mem_table_fill_foreign_vcol_set(
  418. dict_table_t* table);
  419. /** Free the vcol_set from all foreign key constraint on the table.
  420. @param[in,out] table innodb table object. */
  421. void
  422. dict_mem_table_free_foreign_vcol_set(
  423. dict_table_t* table);
  424. /** Create a temporary tablename like "#sql-ibNNN".
  425. @param[in] heap A memory heap
  426. @param[in] dbtab Table name in the form database/table name
  427. @param[in] id Table id
  428. @return A unique temporary tablename suitable for InnoDB use */
  429. char*
  430. dict_mem_create_temporary_tablename(
  431. mem_heap_t* heap,
  432. const char* dbtab,
  433. table_id_t id);
  434. /** SQL identifier name wrapper for pretty-printing */
  435. class id_name_t
  436. {
  437. public:
  438. /** Default constructor */
  439. id_name_t()
  440. : m_name()
  441. {}
  442. /** Constructor
  443. @param[in] name identifier to assign */
  444. explicit id_name_t(
  445. const char* name)
  446. : m_name(name)
  447. {}
  448. /** Assignment operator
  449. @param[in] name identifier to assign */
  450. id_name_t& operator=(
  451. const char* name)
  452. {
  453. m_name = name;
  454. return(*this);
  455. }
  456. /** Implicit type conversion
  457. @return the name */
  458. operator const char*() const
  459. {
  460. return(m_name);
  461. }
  462. /** Explicit type conversion
  463. @return the name */
  464. const char* operator()() const
  465. {
  466. return(m_name);
  467. }
  468. private:
  469. /** The name in internal representation */
  470. const char* m_name;
  471. };
  472. /** Data structure for a column in a table */
  473. struct dict_col_t{
  474. /*----------------------*/
  475. /** The following are copied from dtype_t,
  476. so that all bit-fields can be packed tightly. */
  477. /* @{ */
  478. unsigned prtype:32; /*!< precise type; MySQL data
  479. type, charset code, flags to
  480. indicate nullability,
  481. signedness, whether this is a
  482. binary string, whether this is
  483. a true VARCHAR where MySQL
  484. uses 2 bytes to store the length */
  485. unsigned mtype:8; /*!< main data type */
  486. /* the remaining fields do not affect alphabetical ordering: */
  487. unsigned len:16; /*!< length; for MySQL data this
  488. is field->pack_length(),
  489. except that for a >= 5.0.3
  490. type true VARCHAR this is the
  491. maximum byte length of the
  492. string data (in addition to
  493. the string, MySQL uses 1 or 2
  494. bytes to store the string length) */
  495. unsigned mbminlen:3; /*!< minimum length of a
  496. character, in bytes */
  497. unsigned mbmaxlen:3; /*!< maximum length of a
  498. character, in bytes */
  499. /*----------------------*/
  500. /* End of definitions copied from dtype_t */
  501. /* @} */
  502. unsigned ind:10; /*!< table column position
  503. (starting from 0) */
  504. unsigned ord_part:1; /*!< nonzero if this column
  505. appears in the ordering fields
  506. of an index */
  507. unsigned max_prefix:12; /*!< maximum index prefix length on
  508. this column. Our current max limit is
  509. 3072 (REC_VERSION_56_MAX_INDEX_COL_LEN)
  510. bytes. */
  511. private:
  512. /** Special value of ind for a dropped column */
  513. static const unsigned DROPPED = 1023;
  514. public:
  515. /** Detach a virtual column from an index.
  516. @param index being-freed index */
  517. inline void detach(const dict_index_t &index);
  518. /** Data for instantly added columns */
  519. struct def_t
  520. {
  521. /** original default value of instantly added column */
  522. const void *data;
  523. /** len of data, or UNIV_SQL_DEFAULT if unavailable */
  524. ulint len;
  525. } def_val;
  526. /** Retrieve the column name.
  527. @param table the table of this column */
  528. const char *name(const dict_table_t &table) const;
  529. /** @return whether this is a virtual column */
  530. bool is_virtual() const { return prtype & DATA_VIRTUAL; }
  531. /** @return whether NULL is an allowed value for this column */
  532. bool is_nullable() const { return !(prtype & DATA_NOT_NULL); }
  533. /** @return whether table of this system field is TRX_ID-based */
  534. bool vers_native() const
  535. {
  536. ut_ad(vers_sys_start() || vers_sys_end());
  537. ut_ad(mtype == DATA_INT || mtype == DATA_FIXBINARY);
  538. return mtype == DATA_INT;
  539. }
  540. /** @return whether this user column (not row_start, row_end)
  541. has System Versioning property */
  542. bool is_versioned() const { return !(~prtype & DATA_VERSIONED); }
  543. /** @return whether this is the system version start */
  544. bool vers_sys_start() const
  545. {
  546. return (prtype & DATA_VERSIONED) == DATA_VERS_START;
  547. }
  548. /** @return whether this is the system version end */
  549. bool vers_sys_end() const
  550. {
  551. return (prtype & DATA_VERSIONED) == DATA_VERS_END;
  552. }
  553. /** @return whether this is an instantly-added column */
  554. bool is_added() const
  555. {
  556. DBUG_ASSERT(def_val.len != UNIV_SQL_DEFAULT || !def_val.data);
  557. return def_val.len != UNIV_SQL_DEFAULT;
  558. }
  559. /** Flag the column instantly dropped */
  560. void set_dropped() { ind = DROPPED; }
  561. /** Flag the column instantly dropped.
  562. @param not_null whether the column was NOT NULL
  563. @param len2 whether the length exceeds 255 bytes
  564. @param fixed_len the fixed length in bytes, or 0 */
  565. void set_dropped(bool not_null, bool len2, unsigned fixed)
  566. {
  567. DBUG_ASSERT(!len2 || !fixed);
  568. prtype= not_null ? DATA_NOT_NULL | DATA_BINARY_TYPE : DATA_BINARY_TYPE;
  569. if (fixed)
  570. {
  571. mtype= DATA_FIXBINARY;
  572. len= static_cast<uint16_t>(fixed);
  573. }
  574. else
  575. {
  576. mtype= DATA_BINARY;
  577. len= len2 ? 65535 : 255;
  578. }
  579. mbminlen= mbmaxlen= 0;
  580. ind= DROPPED;
  581. ord_part= 0;
  582. max_prefix= 0;
  583. }
  584. /** @return whether the column was instantly dropped */
  585. bool is_dropped() const { return ind == DROPPED; }
  586. /** @return whether the column was instantly dropped
  587. @param index the clustered index */
  588. inline bool is_dropped(const dict_index_t &index) const;
  589. /** Get the default value of an instantly-added column.
  590. @param[out] len value length (in bytes), or UNIV_SQL_NULL
  591. @return default value
  592. @retval NULL if the default value is SQL NULL (len=UNIV_SQL_NULL) */
  593. const byte *instant_value(ulint *len) const
  594. {
  595. DBUG_ASSERT(is_added());
  596. *len= def_val.len;
  597. return static_cast<const byte*>(def_val.data);
  598. }
  599. /** Remove the 'instant ADD' status of the column */
  600. void clear_instant()
  601. {
  602. def_val.len= UNIV_SQL_DEFAULT;
  603. def_val.data= NULL;
  604. }
  605. /** @return whether two columns have compatible data type encoding */
  606. bool same_type(const dict_col_t &other) const
  607. {
  608. if (mtype != other.mtype)
  609. {
  610. /* For latin1_swedish_ci, DATA_CHAR and DATA_VARCHAR
  611. will be used instead of DATA_MYSQL and DATA_VARMYSQL.
  612. As long as mtype,prtype are being written to InnoDB
  613. data dictionary tables, we cannot simplify this. */
  614. switch (mtype) {
  615. default:
  616. return false;
  617. case DATA_VARCHAR:
  618. if (other.mtype != DATA_VARMYSQL)
  619. return false;
  620. goto check_encoding;
  621. case DATA_VARMYSQL:
  622. if (other.mtype != DATA_VARCHAR)
  623. return false;
  624. goto check_encoding;
  625. case DATA_CHAR:
  626. if (other.mtype != DATA_MYSQL)
  627. return false;
  628. goto check_encoding;
  629. case DATA_MYSQL:
  630. if (other.mtype != DATA_CHAR)
  631. return false;
  632. goto check_encoding;
  633. }
  634. }
  635. else if (dtype_is_string_type(mtype))
  636. {
  637. check_encoding:
  638. const uint16_t cset= dtype_get_charset_coll(prtype);
  639. const uint16_t ocset= dtype_get_charset_coll(other.prtype);
  640. return cset == ocset || dict_col_t::same_encoding(cset, ocset);
  641. }
  642. return true;
  643. }
  644. /** @return whether two collations codes have the same character encoding */
  645. static bool same_encoding(uint16_t a, uint16_t b);
  646. /** Determine if the columns have the same format
  647. except for is_nullable() and is_versioned().
  648. @param other column to compare to
  649. @return whether the columns have the same format */
  650. bool same_format(const dict_col_t &other) const
  651. {
  652. return same_type(other) && len >= other.len &&
  653. mbminlen == other.mbminlen && mbmaxlen >= other.mbmaxlen &&
  654. !((prtype ^ other.prtype) & ~(DATA_NOT_NULL | DATA_VERSIONED |
  655. CHAR_COLL_MASK << 16 |
  656. DATA_LONG_TRUE_VARCHAR));
  657. }
  658. /** @return whether the column values are comparable by memcmp() */
  659. bool is_binary() const { return prtype & DATA_BINARY_TYPE; }
  660. };
  661. /** Index information put in a list of virtual column structure. Index
  662. id and virtual column position in the index will be logged.
  663. There can be multiple entries for a given index, with a different position. */
  664. struct dict_v_idx_t {
  665. /** active index on the column */
  666. dict_index_t* index;
  667. /** position in this index */
  668. ulint nth_field;
  669. dict_v_idx_t(dict_index_t* index, ulint nth_field)
  670. : index(index), nth_field(nth_field) {}
  671. };
  672. /** Data structure for a virtual column in a table */
  673. struct dict_v_col_t{
  674. /** column structure */
  675. dict_col_t m_col;
  676. /** array of base column ptr */
  677. dict_col_t** base_col;
  678. /** number of base column */
  679. unsigned num_base:10;
  680. /** column pos in table */
  681. unsigned v_pos:10;
  682. /** Virtual index list, and column position in the index */
  683. std::forward_list<dict_v_idx_t, ut_allocator<dict_v_idx_t> >
  684. v_indexes;
  685. /** Detach the column from an index.
  686. @param index index to be detached from */
  687. void detach(const dict_index_t &index)
  688. {
  689. if (v_indexes.empty()) return;
  690. auto i= v_indexes.before_begin();
  691. do {
  692. auto prev = i++;
  693. if (i == v_indexes.end())
  694. {
  695. return;
  696. }
  697. if (i->index == &index)
  698. {
  699. v_indexes.erase_after(prev);
  700. return;
  701. }
  702. }
  703. while (i != v_indexes.end());
  704. }
  705. };
  706. /** Data structure for newly added virtual column in a index.
  707. It is used only during rollback_inplace_alter_table() of
  708. addition of index depending on newly added virtual columns
  709. and uses index heap. Should be freed when index is being
  710. removed from cache. */
  711. struct dict_add_v_col_info
  712. {
  713. ulint n_v_col;
  714. dict_v_col_t *v_col;
  715. /** Add the newly added virtual column while rollbacking
  716. the index which contains new virtual columns
  717. @param col virtual column to be duplicated
  718. @param offset offset where to duplicate virtual column */
  719. dict_v_col_t* add_drop_v_col(mem_heap_t *heap, dict_v_col_t *col,
  720. ulint offset)
  721. {
  722. ut_ad(n_v_col);
  723. ut_ad(offset < n_v_col);
  724. if (!v_col)
  725. v_col= static_cast<dict_v_col_t*>
  726. (mem_heap_alloc(heap, n_v_col * sizeof *v_col));
  727. new (&v_col[offset]) dict_v_col_t();
  728. v_col[offset].m_col= col->m_col;
  729. v_col[offset].v_pos= col->v_pos;
  730. return &v_col[offset];
  731. }
  732. };
  733. /** Data structure for newly added virtual column in a table */
  734. struct dict_add_v_col_t{
  735. /** number of new virtual column */
  736. ulint n_v_col;
  737. /** column structures */
  738. const dict_v_col_t* v_col;
  739. /** new col names */
  740. const char** v_col_name;
  741. };
  742. /** Data structure for a stored column in a table. */
  743. struct dict_s_col_t {
  744. /** Stored column ptr */
  745. dict_col_t* m_col;
  746. /** array of base col ptr */
  747. dict_col_t** base_col;
  748. /** number of base columns */
  749. ulint num_base;
  750. /** column pos in table */
  751. ulint s_pos;
  752. };
  753. /** list to put stored column for create_table_info_t */
  754. typedef std::forward_list<dict_s_col_t, ut_allocator<dict_s_col_t> >
  755. dict_s_col_list;
  756. /** @brief DICT_ANTELOPE_MAX_INDEX_COL_LEN is measured in bytes and
  757. is the maximum indexed column length (or indexed prefix length) in
  758. ROW_FORMAT=REDUNDANT and ROW_FORMAT=COMPACT. Also, in any format,
  759. any fixed-length field that is longer than this will be encoded as
  760. a variable-length field.
  761. It is set to 3*256, so that one can create a column prefix index on
  762. 256 characters of a TEXT or VARCHAR column also in the UTF-8
  763. charset. In that charset, a character may take at most 3 bytes. This
  764. constant MUST NOT BE CHANGED, or the compatibility of InnoDB data
  765. files would be at risk! */
  766. #define DICT_ANTELOPE_MAX_INDEX_COL_LEN REC_ANTELOPE_MAX_INDEX_COL_LEN
  767. /** Find out maximum indexed column length by its table format.
  768. For ROW_FORMAT=REDUNDANT and ROW_FORMAT=COMPACT, the maximum
  769. field length is REC_ANTELOPE_MAX_INDEX_COL_LEN - 1 (767). For
  770. ROW_FORMAT=COMPRESSED and ROW_FORMAT=DYNAMIC, the length could
  771. be REC_VERSION_56_MAX_INDEX_COL_LEN (3072) bytes */
  772. #define DICT_MAX_FIELD_LEN_BY_FORMAT(table) \
  773. (dict_table_has_atomic_blobs(table) \
  774. ? REC_VERSION_56_MAX_INDEX_COL_LEN \
  775. : REC_ANTELOPE_MAX_INDEX_COL_LEN - 1)
  776. #define DICT_MAX_FIELD_LEN_BY_FORMAT_FLAG(flags) \
  777. (DICT_TF_HAS_ATOMIC_BLOBS(flags) \
  778. ? REC_VERSION_56_MAX_INDEX_COL_LEN \
  779. : REC_ANTELOPE_MAX_INDEX_COL_LEN - 1)
  780. /** Defines the maximum fixed length column size */
  781. #define DICT_MAX_FIXED_COL_LEN DICT_ANTELOPE_MAX_INDEX_COL_LEN
  782. #ifdef WITH_WSREP
  783. #define WSREP_MAX_SUPPORTED_KEY_LENGTH 3500
  784. #endif /* WITH_WSREP */
  785. /** Data structure for a field in an index */
  786. struct dict_field_t{
  787. dict_col_t* col; /*!< pointer to the table column */
  788. id_name_t name; /*!< name of the column */
  789. unsigned prefix_len:12; /*!< 0 or the length of the column
  790. prefix in bytes in a MySQL index of
  791. type, e.g., INDEX (textcol(25));
  792. must be smaller than
  793. DICT_MAX_FIELD_LEN_BY_FORMAT;
  794. NOTE that in the UTF-8 charset, MySQL
  795. sets this to (mbmaxlen * the prefix len)
  796. in UTF-8 chars */
  797. unsigned fixed_len:10; /*!< 0 or the fixed length of the
  798. column if smaller than
  799. DICT_ANTELOPE_MAX_INDEX_COL_LEN */
  800. /** Zero-initialize all fields */
  801. dict_field_t() : col(NULL), name(NULL), prefix_len(0), fixed_len(0) {}
  802. /** Check whether two index fields are equivalent.
  803. @param[in] old the other index field
  804. @return whether the index fields are equivalent */
  805. bool same(const dict_field_t& other) const
  806. {
  807. return(prefix_len == other.prefix_len
  808. && fixed_len == other.fixed_len);
  809. }
  810. };
  811. /**********************************************************************//**
  812. PADDING HEURISTIC BASED ON LINEAR INCREASE OF PADDING TO AVOID
  813. COMPRESSION FAILURES
  814. (Note: this is relevant only for compressed indexes)
  815. GOAL: Avoid compression failures by maintaining information about the
  816. compressibility of data. If data is not very compressible then leave
  817. some extra space 'padding' in the uncompressed page making it more
  818. likely that compression of less than fully packed uncompressed page will
  819. succeed.
  820. This padding heuristic works by increasing the pad linearly until the
  821. desired failure rate is reached. A "round" is a fixed number of
  822. compression operations.
  823. After each round, the compression failure rate for that round is
  824. computed. If the failure rate is too high, then padding is incremented
  825. by a fixed value, otherwise it's left intact.
  826. If the compression failure is lower than the desired rate for a fixed
  827. number of consecutive rounds, then the padding is decreased by a fixed
  828. value. This is done to prevent overshooting the padding value,
  829. and to accommodate the possible change in data compressibility. */
  830. /** Number of zip ops in one round. */
  831. #define ZIP_PAD_ROUND_LEN (128)
  832. /** Number of successful rounds after which the padding is decreased */
  833. #define ZIP_PAD_SUCCESSFUL_ROUND_LIMIT (5)
  834. /** Amount by which padding is increased. */
  835. #define ZIP_PAD_INCR (128)
  836. /** Percentage of compression failures that are allowed in a single
  837. round */
  838. extern ulong zip_failure_threshold_pct;
  839. /** Maximum percentage of a page that can be allowed as a pad to avoid
  840. compression failures */
  841. extern ulong zip_pad_max;
  842. /** Data structure to hold information about about how much space in
  843. an uncompressed page should be left as padding to avoid compression
  844. failures. This estimate is based on a self-adapting heuristic. */
  845. struct zip_pad_info_t {
  846. /** Dummy assignment operator for dict_index_t::clone() */
  847. zip_pad_info_t &operator=(const zip_pad_info_t&) { return *this; }
  848. std::mutex mutex; /*!< mutex protecting the info */
  849. Atomic_relaxed<ulint>
  850. pad; /*!< number of bytes used as pad */
  851. ulint success;/*!< successful compression ops during
  852. current round */
  853. ulint failure;/*!< failed compression ops during
  854. current round */
  855. ulint n_rounds;/*!< number of currently successful
  856. rounds */
  857. };
  858. /** Number of samples of data size kept when page compression fails for
  859. a certain index.*/
  860. #define STAT_DEFRAG_DATA_SIZE_N_SAMPLE 10
  861. /** "GEN_CLUST_INDEX" is the name reserved for InnoDB default
  862. system clustered index when there is no primary key. */
  863. const char innobase_index_reserve_name[] = "GEN_CLUST_INDEX";
  864. /** Data structure for an index. Most fields will be
  865. initialized to 0, NULL or FALSE in dict_mem_index_create(). */
  866. struct dict_index_t {
  867. /** Maximum number of fields */
  868. static constexpr unsigned MAX_N_FIELDS= (1U << 10) - 1;
  869. index_id_t id; /*!< id of the index */
  870. mem_heap_t* heap; /*!< memory heap */
  871. id_name_t name; /*!< index name */
  872. dict_table_t* table; /*!< back pointer to table */
  873. /** root page number, or FIL_NULL if the index has been detached
  874. from storage (DISCARD TABLESPACE or similar),
  875. or 1 if the index is in table->freed_indexes */
  876. unsigned page:32;
  877. unsigned merge_threshold:6;
  878. /*!< In the pessimistic delete, if the page
  879. data size drops below this limit in percent,
  880. merging it to a neighbor is tried */
  881. # define DICT_INDEX_MERGE_THRESHOLD_DEFAULT 50
  882. unsigned type:DICT_IT_BITS;
  883. /*!< index type (DICT_CLUSTERED, DICT_UNIQUE,
  884. DICT_IBUF, DICT_CORRUPT) */
  885. #define MAX_KEY_LENGTH_BITS 12
  886. unsigned trx_id_offset:MAX_KEY_LENGTH_BITS;
  887. /*!< position of the trx id column
  888. in a clustered index record, if the fields
  889. before it are known to be of a fixed size,
  890. 0 otherwise */
  891. #if (1<<MAX_KEY_LENGTH_BITS) < HA_MAX_KEY_LENGTH
  892. # error (1<<MAX_KEY_LENGTH_BITS) < HA_MAX_KEY_LENGTH
  893. #endif
  894. unsigned n_user_defined_cols:10;
  895. /*!< number of columns the user defined to
  896. be in the index: in the internal
  897. representation we add more columns */
  898. unsigned nulls_equal:1;
  899. /*!< if true, SQL NULL == SQL NULL */
  900. #ifdef BTR_CUR_HASH_ADAPT
  901. #ifdef MYSQL_INDEX_DISABLE_AHI
  902. unsigned disable_ahi:1;
  903. /*!< whether to disable the
  904. adaptive hash index.
  905. Maybe this could be disabled for
  906. temporary tables? */
  907. #endif
  908. #endif /* BTR_CUR_HASH_ADAPT */
  909. unsigned n_uniq:10;/*!< number of fields from the beginning
  910. which are enough to determine an index
  911. entry uniquely */
  912. unsigned n_def:10;/*!< number of fields defined so far */
  913. unsigned n_fields:10;/*!< number of fields in the index */
  914. unsigned n_nullable:10;/*!< number of nullable fields */
  915. unsigned n_core_fields:10;/*!< number of fields in the index
  916. (before the first time of instant add columns) */
  917. /** number of bytes of null bits in ROW_FORMAT!=REDUNDANT node pointer
  918. records; usually equal to UT_BITS_IN_BYTES(n_nullable), but
  919. can be less in clustered indexes with instant ADD COLUMN */
  920. unsigned n_core_null_bytes:8;
  921. /** magic value signalling that n_core_null_bytes was not
  922. initialized yet */
  923. static const unsigned NO_CORE_NULL_BYTES = 0xff;
  924. /** The clustered index ID of the hard-coded SYS_INDEXES table. */
  925. static const unsigned DICT_INDEXES_ID = 3;
  926. unsigned cached:1;/*!< TRUE if the index object is in the
  927. dictionary cache */
  928. unsigned to_be_dropped:1;
  929. /*!< TRUE if the index is to be dropped;
  930. protected by dict_sys.latch */
  931. unsigned online_status:2;
  932. /*!< enum online_index_status.
  933. Transitions from ONLINE_INDEX_COMPLETE (to
  934. ONLINE_INDEX_CREATION) are protected
  935. by dict_sys.latch and
  936. dict_sys.mutex. Other changes are
  937. protected by index->lock. */
  938. unsigned uncommitted:1;
  939. /*!< a flag that is set for secondary indexes
  940. that have not been committed to the
  941. data dictionary yet */
  942. #ifdef UNIV_DEBUG
  943. /** whether this is a dummy index object */
  944. bool is_dummy;
  945. /** whether btr_cur_instant_init() is in progress */
  946. bool in_instant_init;
  947. uint32_t magic_n;/*!< magic number */
  948. /** Value of dict_index_t::magic_n */
  949. # define DICT_INDEX_MAGIC_N 76789786
  950. #endif
  951. dict_field_t* fields; /*!< array of field descriptions */
  952. st_mysql_ftparser*
  953. parser; /*!< fulltext parser plugin */
  954. /** It just indicates whether newly added virtual column
  955. during alter. It stores column in case of alter failure.
  956. It should use heap from dict_index_t. It should be freed
  957. while removing the index from table. */
  958. dict_add_v_col_info* new_vcol_info;
  959. UT_LIST_NODE_T(dict_index_t)
  960. indexes;/*!< list of indexes of the table */
  961. #ifdef BTR_CUR_ADAPT
  962. btr_search_t* search_info;
  963. /*!< info used in optimistic searches */
  964. #endif /* BTR_CUR_ADAPT */
  965. row_log_t* online_log;
  966. /*!< the log of modifications
  967. during online index creation;
  968. valid when online_status is
  969. ONLINE_INDEX_CREATION */
  970. /*----------------------*/
  971. /** Statistics for query optimization */
  972. /* @{ */
  973. ib_uint64_t* stat_n_diff_key_vals;
  974. /*!< approximate number of different
  975. key values for this index, for each
  976. n-column prefix where 1 <= n <=
  977. dict_get_n_unique(index) (the array is
  978. indexed from 0 to n_uniq-1); we
  979. periodically calculate new
  980. estimates */
  981. ib_uint64_t* stat_n_sample_sizes;
  982. /*!< number of pages that were sampled
  983. to calculate each of stat_n_diff_key_vals[],
  984. e.g. stat_n_sample_sizes[3] pages were sampled
  985. to get the number stat_n_diff_key_vals[3]. */
  986. ib_uint64_t* stat_n_non_null_key_vals;
  987. /* approximate number of non-null key values
  988. for this index, for each column where
  989. 1 <= n <= dict_get_n_unique(index) (the array
  990. is indexed from 0 to n_uniq-1); This
  991. is used when innodb_stats_method is
  992. "nulls_ignored". */
  993. ulint stat_index_size;
  994. /*!< approximate index size in
  995. database pages */
  996. ulint stat_n_leaf_pages;
  997. /*!< approximate number of leaf pages in the
  998. index tree */
  999. bool stats_error_printed;
  1000. /*!< has persistent statistics error printed
  1001. for this index ? */
  1002. /* @} */
  1003. /** Statistics for defragmentation, these numbers are estimations and
  1004. could be very inaccurate at certain times, e.g. right after restart,
  1005. during defragmentation, etc. */
  1006. /* @{ */
  1007. ulint stat_defrag_modified_counter;
  1008. ulint stat_defrag_n_pages_freed;
  1009. /* number of pages freed by defragmentation. */
  1010. ulint stat_defrag_n_page_split;
  1011. /* number of page splits since last full index
  1012. defragmentation. */
  1013. ulint stat_defrag_data_size_sample[STAT_DEFRAG_DATA_SIZE_N_SAMPLE];
  1014. /* data size when compression failure happened
  1015. the most recent 10 times. */
  1016. ulint stat_defrag_sample_next_slot;
  1017. /* in which slot the next sample should be
  1018. saved. */
  1019. /* @} */
  1020. private:
  1021. /** R-tree split sequence number */
  1022. Atomic_relaxed<node_seq_t> rtr_ssn;
  1023. public:
  1024. void set_ssn(node_seq_t ssn) { rtr_ssn= ssn; }
  1025. node_seq_t assign_ssn() { return rtr_ssn.fetch_add(1) + 1; }
  1026. node_seq_t ssn() const { return rtr_ssn; }
  1027. rtr_info_track_t*
  1028. rtr_track;/*!< tracking all R-Tree search cursors */
  1029. trx_id_t trx_id; /*!< id of the transaction that created this
  1030. index, or 0 if the index existed
  1031. when InnoDB was started up */
  1032. zip_pad_info_t zip_pad;/*!< Information about state of
  1033. compression failures and successes */
  1034. mutable rw_lock_t lock; /*!< read-write lock protecting the
  1035. upper levels of the index tree */
  1036. /** Determine if the index has been committed to the
  1037. data dictionary.
  1038. @return whether the index definition has been committed */
  1039. bool is_committed() const
  1040. {
  1041. ut_ad(!uncommitted || !(type & DICT_CLUSTERED));
  1042. return(UNIV_LIKELY(!uncommitted));
  1043. }
  1044. /** Flag an index committed or uncommitted.
  1045. @param[in] committed whether the index is committed */
  1046. void set_committed(bool committed)
  1047. {
  1048. ut_ad(!to_be_dropped);
  1049. ut_ad(committed || !(type & DICT_CLUSTERED));
  1050. uncommitted = !committed;
  1051. }
  1052. /** Notify that the index pages are going to be modified.
  1053. @param[in,out] mtr mini-transaction */
  1054. inline void set_modified(mtr_t& mtr) const;
  1055. /** @return whether this index is readable
  1056. @retval true normally
  1057. @retval false if this is a single-table tablespace
  1058. and the .ibd file is missing, or a
  1059. page cannot be read or decrypted */
  1060. inline bool is_readable() const;
  1061. /** @return whether instant ALTER TABLE is in effect */
  1062. inline bool is_instant() const;
  1063. /** @return whether the index is the primary key index
  1064. (not the clustered index of the change buffer) */
  1065. bool is_primary() const
  1066. {
  1067. return DICT_CLUSTERED == (type & (DICT_CLUSTERED | DICT_IBUF));
  1068. }
  1069. /** @return whether this is a generated clustered index */
  1070. bool is_gen_clust() const { return type == DICT_CLUSTERED; }
  1071. /** @return whether this is a clustered index */
  1072. bool is_clust() const { return type & DICT_CLUSTERED; }
  1073. /** @return whether this is a unique index */
  1074. bool is_unique() const { return type & DICT_UNIQUE; }
  1075. /** @return whether this is a spatial index */
  1076. bool is_spatial() const { return UNIV_UNLIKELY(type & DICT_SPATIAL); }
  1077. /** @return whether this is the change buffer */
  1078. bool is_ibuf() const { return UNIV_UNLIKELY(type & DICT_IBUF); }
  1079. /** @return whether the index includes virtual columns */
  1080. bool has_virtual() const { return type & DICT_VIRTUAL; }
  1081. /** @return the position of DB_TRX_ID */
  1082. uint16_t db_trx_id() const {
  1083. DBUG_ASSERT(is_primary());
  1084. DBUG_ASSERT(n_uniq);
  1085. DBUG_ASSERT(n_uniq <= MAX_REF_PARTS);
  1086. return n_uniq;
  1087. }
  1088. /** @return the position of DB_ROLL_PTR */
  1089. uint16_t db_roll_ptr() const
  1090. {
  1091. return static_cast<uint16_t>(db_trx_id() + 1);
  1092. }
  1093. /** @return the offset of the metadata BLOB field,
  1094. or the first user field after the PRIMARY KEY,DB_TRX_ID,DB_ROLL_PTR */
  1095. uint16_t first_user_field() const
  1096. {
  1097. return static_cast<uint16_t>(db_trx_id() + 2);
  1098. }
  1099. /** @return whether the index is corrupted */
  1100. inline bool is_corrupted() const;
  1101. /** Detach the virtual columns from the index that is to be removed. */
  1102. void detach_columns()
  1103. {
  1104. if (!has_virtual() || !cached)
  1105. return;
  1106. for (unsigned i= 0; i < n_fields; i++)
  1107. {
  1108. dict_col_t* col= fields[i].col;
  1109. if (!col || !col->is_virtual())
  1110. continue;
  1111. col->detach(*this);
  1112. }
  1113. }
  1114. /** Determine how many fields of a given prefix can be set NULL.
  1115. @param[in] n_prefix number of fields in the prefix
  1116. @return number of fields 0..n_prefix-1 that can be set NULL */
  1117. unsigned get_n_nullable(ulint n_prefix) const
  1118. {
  1119. DBUG_ASSERT(n_prefix > 0);
  1120. DBUG_ASSERT(n_prefix <= n_fields);
  1121. unsigned n = n_nullable;
  1122. for (; n_prefix < n_fields; n_prefix++) {
  1123. const dict_col_t* col = fields[n_prefix].col;
  1124. DBUG_ASSERT(!col->is_virtual());
  1125. n -= col->is_nullable();
  1126. }
  1127. DBUG_ASSERT(n < n_def);
  1128. return n;
  1129. }
  1130. /** Get the default value of an instantly-added clustered index field.
  1131. @param[in] n instantly added field position
  1132. @param[out] len value length (in bytes), or UNIV_SQL_NULL
  1133. @return default value
  1134. @retval NULL if the default value is SQL NULL (len=UNIV_SQL_NULL) */
  1135. const byte* instant_field_value(ulint n, ulint* len) const
  1136. {
  1137. DBUG_ASSERT(is_instant() || id == DICT_INDEXES_ID);
  1138. DBUG_ASSERT(n + (id == DICT_INDEXES_ID) >= n_core_fields);
  1139. DBUG_ASSERT(n < n_fields);
  1140. return fields[n].col->instant_value(len);
  1141. }
  1142. /** Adjust index metadata for instant ADD/DROP/reorder COLUMN.
  1143. @param[in] clustered index definition after instant ALTER TABLE */
  1144. inline void instant_add_field(const dict_index_t& instant);
  1145. /** Remove instant ADD COLUMN metadata. */
  1146. inline void clear_instant_add();
  1147. /** Remove instant ALTER TABLE metadata. */
  1148. inline void clear_instant_alter();
  1149. /** Construct the metadata record for instant ALTER TABLE.
  1150. @param[in] row dummy or default values for existing columns
  1151. @param[in,out] heap memory heap for allocations
  1152. @return metadata record */
  1153. inline dtuple_t*
  1154. instant_metadata(const dtuple_t& row, mem_heap_t* heap) const;
  1155. /** Check if record in clustered index is historical row.
  1156. @param[in] rec clustered row
  1157. @param[in] offsets offsets
  1158. @return true if row is historical */
  1159. bool
  1160. vers_history_row(const rec_t* rec, const rec_offs* offsets);
  1161. /** Check if record in secondary index is historical row.
  1162. @param[in] rec record in a secondary index
  1163. @param[out] history_row true if row is historical
  1164. @return true on error */
  1165. bool
  1166. vers_history_row(const rec_t* rec, bool &history_row);
  1167. /** Assign the number of new column to be added as a part
  1168. of the index
  1169. @param n_vcol number of virtual columns to be added */
  1170. void assign_new_v_col(ulint n_vcol)
  1171. {
  1172. new_vcol_info= static_cast<dict_add_v_col_info*>
  1173. (mem_heap_zalloc(heap, sizeof *new_vcol_info));
  1174. new_vcol_info->n_v_col= n_vcol;
  1175. }
  1176. /* @return whether index has new virtual column */
  1177. bool has_new_v_col() const { return new_vcol_info; }
  1178. /* @return number of newly added virtual column */
  1179. ulint get_new_n_vcol() const
  1180. { return new_vcol_info ? new_vcol_info->n_v_col : 0; }
  1181. /** Reconstruct the clustered index fields.
  1182. @return whether metadata is incorrect */
  1183. inline bool reconstruct_fields();
  1184. /** Check if the index contains a column or a prefix of that column.
  1185. @param[in] n column number
  1186. @param[in] is_virtual whether it is a virtual col
  1187. @return whether the index contains the column or its prefix */
  1188. bool contains_col_or_prefix(ulint n, bool is_virtual) const
  1189. MY_ATTRIBUTE((warn_unused_result));
  1190. #ifdef BTR_CUR_HASH_ADAPT
  1191. /** @return a clone of this */
  1192. dict_index_t* clone() const;
  1193. /** Clone this index for lazy dropping of the adaptive hash index.
  1194. @return this or a clone */
  1195. dict_index_t* clone_if_needed();
  1196. /** @return number of leaf pages pointed to by the adaptive hash index */
  1197. inline ulint n_ahi_pages() const;
  1198. /** @return whether mark_freed() had been invoked */
  1199. bool freed() const { return UNIV_UNLIKELY(page == 1); }
  1200. /** Note that the index is waiting for btr_search_lazy_free() */
  1201. void set_freed() { ut_ad(!freed()); page= 1; }
  1202. #endif /* BTR_CUR_HASH_ADAPT */
  1203. /** @return whether it is forbidden to invoke clear_instant_add() */
  1204. bool must_avoid_clear_instant_add() const
  1205. {
  1206. if (is_instant())
  1207. for (auto i= this; (i= UT_LIST_GET_NEXT(indexes, i)) != nullptr; )
  1208. if (i->to_be_dropped /* || i->online_log*/)
  1209. return true;
  1210. return false;
  1211. }
  1212. /** This ad-hoc class is used by record_size_info only. */
  1213. class record_size_info_t {
  1214. public:
  1215. record_size_info_t()
  1216. : max_leaf_size(0), shortest_size(0), too_big(false),
  1217. first_overrun_field_index(SIZE_T_MAX), overrun_size(0)
  1218. {
  1219. }
  1220. /** Mark row potentially too big for page and set up first
  1221. overflow field index. */
  1222. void set_too_big(size_t field_index)
  1223. {
  1224. ut_ad(field_index != SIZE_T_MAX);
  1225. too_big = true;
  1226. if (first_overrun_field_index > field_index) {
  1227. first_overrun_field_index = field_index;
  1228. overrun_size = shortest_size;
  1229. }
  1230. }
  1231. /** @return overrun field index or SIZE_T_MAX if nothing
  1232. overflowed*/
  1233. size_t get_first_overrun_field_index() const
  1234. {
  1235. ut_ad(row_is_too_big());
  1236. ut_ad(first_overrun_field_index != SIZE_T_MAX);
  1237. return first_overrun_field_index;
  1238. }
  1239. size_t get_overrun_size() const
  1240. {
  1241. ut_ad(row_is_too_big());
  1242. return overrun_size;
  1243. }
  1244. bool row_is_too_big() const { return too_big; }
  1245. size_t max_leaf_size; /** Bigger row size this index can
  1246. produce */
  1247. size_t shortest_size; /** shortest because it counts everything
  1248. as in overflow pages */
  1249. private:
  1250. bool too_big; /** This one is true when maximum row size this
  1251. index can produce is bigger than maximum row
  1252. size given page can hold. */
  1253. size_t first_overrun_field_index; /** After adding this field
  1254. index row overflowed maximum
  1255. allowed size. Useful for
  1256. reporting back to user. */
  1257. size_t overrun_size; /** Just overrun row size */
  1258. };
  1259. /** Returns max possibly record size for that index, size of a shortest
  1260. everything in overflow) size of the longest possible row and index
  1261. of a field which made index records too big to fit on a page.*/
  1262. inline record_size_info_t record_size_info() const;
  1263. };
  1264. /** Detach a virtual column from an index.
  1265. @param index being-freed index */
  1266. inline void dict_col_t::detach(const dict_index_t &index)
  1267. {
  1268. if (is_virtual())
  1269. reinterpret_cast<dict_v_col_t*>(this)->detach(index);
  1270. }
  1271. /** The status of online index creation */
  1272. enum online_index_status {
  1273. /** the index is complete and ready for access */
  1274. ONLINE_INDEX_COMPLETE = 0,
  1275. /** the index is being created, online
  1276. (allowing concurrent modifications) */
  1277. ONLINE_INDEX_CREATION,
  1278. /** secondary index creation was aborted and the index
  1279. should be dropped as soon as index->table->n_ref_count reaches 0,
  1280. or online table rebuild was aborted and the clustered index
  1281. of the original table should soon be restored to
  1282. ONLINE_INDEX_COMPLETE */
  1283. ONLINE_INDEX_ABORTED,
  1284. /** the online index creation was aborted, the index was
  1285. dropped from the data dictionary and the tablespace, and it
  1286. should be dropped from the data dictionary cache as soon as
  1287. index->table->n_ref_count reaches 0. */
  1288. ONLINE_INDEX_ABORTED_DROPPED
  1289. };
  1290. /** Set to store the virtual columns which are affected by Foreign
  1291. key constraint. */
  1292. typedef std::set<dict_v_col_t*, std::less<dict_v_col_t*>,
  1293. ut_allocator<dict_v_col_t*> > dict_vcol_set;
  1294. /** Data structure for a foreign key constraint; an example:
  1295. FOREIGN KEY (A, B) REFERENCES TABLE2 (C, D). Most fields will be
  1296. initialized to 0, NULL or FALSE in dict_mem_foreign_create(). */
  1297. struct dict_foreign_t{
  1298. mem_heap_t* heap; /*!< this object is allocated from
  1299. this memory heap */
  1300. char* id; /*!< id of the constraint as a
  1301. null-terminated string */
  1302. unsigned n_fields:10; /*!< number of indexes' first fields
  1303. for which the foreign key
  1304. constraint is defined: we allow the
  1305. indexes to contain more fields than
  1306. mentioned in the constraint, as long
  1307. as the first fields are as mentioned */
  1308. unsigned type:6; /*!< 0 or DICT_FOREIGN_ON_DELETE_CASCADE
  1309. or DICT_FOREIGN_ON_DELETE_SET_NULL */
  1310. char* foreign_table_name;/*!< foreign table name */
  1311. char* foreign_table_name_lookup;
  1312. /*!< foreign table name used for dict lookup */
  1313. dict_table_t* foreign_table; /*!< table where the foreign key is */
  1314. const char** foreign_col_names;/*!< names of the columns in the
  1315. foreign key */
  1316. char* referenced_table_name;/*!< referenced table name */
  1317. char* referenced_table_name_lookup;
  1318. /*!< referenced table name for dict lookup*/
  1319. dict_table_t* referenced_table;/*!< table where the referenced key
  1320. is */
  1321. const char** referenced_col_names;/*!< names of the referenced
  1322. columns in the referenced table */
  1323. dict_index_t* foreign_index; /*!< foreign index; we require that
  1324. both tables contain explicitly defined
  1325. indexes for the constraint: InnoDB
  1326. does not generate new indexes
  1327. implicitly */
  1328. dict_index_t* referenced_index;/*!< referenced index */
  1329. dict_vcol_set* v_cols; /*!< set of virtual columns affected
  1330. by foreign key constraint. */
  1331. /** Check whether the fulltext index gets affected by
  1332. foreign key constraint */
  1333. bool affects_fulltext() const;
  1334. };
  1335. std::ostream&
  1336. operator<< (std::ostream& out, const dict_foreign_t& foreign);
  1337. struct dict_foreign_print {
  1338. dict_foreign_print(std::ostream& out)
  1339. : m_out(out)
  1340. {}
  1341. void operator()(const dict_foreign_t* foreign) {
  1342. m_out << *foreign;
  1343. }
  1344. private:
  1345. std::ostream& m_out;
  1346. };
  1347. /** Compare two dict_foreign_t objects using their ids. Used in the ordering
  1348. of dict_table_t::foreign_set and dict_table_t::referenced_set. It returns
  1349. true if the first argument is considered to go before the second in the
  1350. strict weak ordering it defines, and false otherwise. */
  1351. struct dict_foreign_compare {
  1352. bool operator()(
  1353. const dict_foreign_t* lhs,
  1354. const dict_foreign_t* rhs) const
  1355. {
  1356. return strcmp(lhs->id, rhs->id) < 0;
  1357. }
  1358. };
  1359. /** A function object to find a foreign key with the given index as the
  1360. referenced index. Return the foreign key with matching criteria or NULL */
  1361. struct dict_foreign_with_index {
  1362. dict_foreign_with_index(const dict_index_t* index)
  1363. : m_index(index)
  1364. {}
  1365. bool operator()(const dict_foreign_t* foreign) const
  1366. {
  1367. return(foreign->referenced_index == m_index);
  1368. }
  1369. const dict_index_t* m_index;
  1370. };
  1371. #ifdef WITH_WSREP
  1372. /** A function object to find a foreign key with the given index as the
  1373. foreign index. Return the foreign key with matching criteria or NULL */
  1374. struct dict_foreign_with_foreign_index {
  1375. dict_foreign_with_foreign_index(const dict_index_t* index)
  1376. : m_index(index)
  1377. {}
  1378. bool operator()(const dict_foreign_t* foreign) const
  1379. {
  1380. return(foreign->foreign_index == m_index);
  1381. }
  1382. const dict_index_t* m_index;
  1383. };
  1384. #endif
  1385. /* A function object to check if the foreign constraint is between different
  1386. tables. Returns true if foreign key constraint is between different tables,
  1387. false otherwise. */
  1388. struct dict_foreign_different_tables {
  1389. bool operator()(const dict_foreign_t* foreign) const
  1390. {
  1391. return(foreign->foreign_table != foreign->referenced_table);
  1392. }
  1393. };
  1394. /** A function object to check if the foreign key constraint has the same
  1395. name as given. If the full name of the foreign key constraint doesn't match,
  1396. then, check if removing the database name from the foreign key constraint
  1397. matches. Return true if it matches, false otherwise. */
  1398. struct dict_foreign_matches_id {
  1399. dict_foreign_matches_id(const char* id)
  1400. : m_id(id)
  1401. {}
  1402. bool operator()(const dict_foreign_t* foreign) const
  1403. {
  1404. if (0 == innobase_strcasecmp(foreign->id, m_id)) {
  1405. return(true);
  1406. }
  1407. if (const char* pos = strchr(foreign->id, '/')) {
  1408. if (0 == innobase_strcasecmp(m_id, pos + 1)) {
  1409. return(true);
  1410. }
  1411. }
  1412. return(false);
  1413. }
  1414. const char* m_id;
  1415. };
  1416. typedef std::set<
  1417. dict_foreign_t*,
  1418. dict_foreign_compare,
  1419. ut_allocator<dict_foreign_t*> > dict_foreign_set;
  1420. std::ostream&
  1421. operator<< (std::ostream& out, const dict_foreign_set& fk_set);
  1422. /** Function object to check if a foreign key object is there
  1423. in the given foreign key set or not. It returns true if the
  1424. foreign key is not found, false otherwise */
  1425. struct dict_foreign_not_exists {
  1426. dict_foreign_not_exists(const dict_foreign_set& obj_)
  1427. : m_foreigns(obj_)
  1428. {}
  1429. /* Return true if the given foreign key is not found */
  1430. bool operator()(dict_foreign_t* const & foreign) const {
  1431. return(m_foreigns.find(foreign) == m_foreigns.end());
  1432. }
  1433. private:
  1434. const dict_foreign_set& m_foreigns;
  1435. };
  1436. /** Validate the search order in the foreign key set.
  1437. @param[in] fk_set the foreign key set to be validated
  1438. @return true if search order is fine in the set, false otherwise. */
  1439. bool
  1440. dict_foreign_set_validate(
  1441. const dict_foreign_set& fk_set);
  1442. /** Validate the search order in the foreign key sets of the table
  1443. (foreign_set and referenced_set).
  1444. @param[in] table table whose foreign key sets are to be validated
  1445. @return true if foreign key sets are fine, false otherwise. */
  1446. bool
  1447. dict_foreign_set_validate(
  1448. const dict_table_t& table);
  1449. /*********************************************************************//**
  1450. Frees a foreign key struct. */
  1451. inline
  1452. void
  1453. dict_foreign_free(
  1454. /*==============*/
  1455. dict_foreign_t* foreign) /*!< in, own: foreign key struct */
  1456. {
  1457. if (foreign->v_cols != NULL) {
  1458. UT_DELETE(foreign->v_cols);
  1459. }
  1460. mem_heap_free(foreign->heap);
  1461. }
  1462. /** The destructor will free all the foreign key constraints in the set
  1463. by calling dict_foreign_free() on each of the foreign key constraints.
  1464. This is used to free the allocated memory when a local set goes out
  1465. of scope. */
  1466. struct dict_foreign_set_free {
  1467. dict_foreign_set_free(const dict_foreign_set& foreign_set)
  1468. : m_foreign_set(foreign_set)
  1469. {}
  1470. ~dict_foreign_set_free()
  1471. {
  1472. std::for_each(m_foreign_set.begin(),
  1473. m_foreign_set.end(),
  1474. dict_foreign_free);
  1475. }
  1476. const dict_foreign_set& m_foreign_set;
  1477. };
  1478. /** The flags for ON_UPDATE and ON_DELETE can be ORed; the default is that
  1479. a foreign key constraint is enforced, therefore RESTRICT just means no flag */
  1480. /* @{ */
  1481. #define DICT_FOREIGN_ON_DELETE_CASCADE 1U /*!< ON DELETE CASCADE */
  1482. #define DICT_FOREIGN_ON_DELETE_SET_NULL 2U /*!< ON UPDATE SET NULL */
  1483. #define DICT_FOREIGN_ON_UPDATE_CASCADE 4U /*!< ON DELETE CASCADE */
  1484. #define DICT_FOREIGN_ON_UPDATE_SET_NULL 8U /*!< ON UPDATE SET NULL */
  1485. #define DICT_FOREIGN_ON_DELETE_NO_ACTION 16U /*!< ON DELETE NO ACTION */
  1486. #define DICT_FOREIGN_ON_UPDATE_NO_ACTION 32U /*!< ON UPDATE NO ACTION */
  1487. /* @} */
  1488. /** Display an identifier.
  1489. @param[in,out] s output stream
  1490. @param[in] id_name SQL identifier (other than table name)
  1491. @return the output stream */
  1492. std::ostream&
  1493. operator<<(
  1494. std::ostream& s,
  1495. const id_name_t& id_name);
  1496. /** Display a table name.
  1497. @param[in,out] s output stream
  1498. @param[in] table_name table name
  1499. @return the output stream */
  1500. std::ostream&
  1501. operator<<(
  1502. std::ostream& s,
  1503. const table_name_t& table_name);
  1504. /** List of locks that different transactions have acquired on a table. This
  1505. list has a list node that is embedded in a nested union/structure. We have to
  1506. generate a specific template for it. */
  1507. typedef ut_list_base<lock_t, ut_list_node<lock_t> lock_table_t::*>
  1508. table_lock_list_t;
  1509. /** mysql template structure defined in row0mysql.cc */
  1510. struct mysql_row_templ_t;
  1511. /** Structure defines template related to virtual columns and
  1512. their base columns */
  1513. struct dict_vcol_templ_t {
  1514. /** number of regular columns */
  1515. ulint n_col;
  1516. /** number of virtual columns */
  1517. ulint n_v_col;
  1518. /** array of templates for virtual col and their base columns */
  1519. mysql_row_templ_t** vtempl;
  1520. /** table's database name */
  1521. std::string db_name;
  1522. /** table name */
  1523. std::string tb_name;
  1524. /** MySQL record length */
  1525. ulint rec_len;
  1526. /** default column value if any */
  1527. byte* default_rec;
  1528. /** cached MySQL TABLE object */
  1529. TABLE* mysql_table;
  1530. /** when mysql_table was cached */
  1531. uint64_t mysql_table_query_id;
  1532. dict_vcol_templ_t() : vtempl(0), mysql_table_query_id(~0ULL) {}
  1533. };
  1534. /** Metadata on clustered index fields starting from first_user_field() */
  1535. class field_map_element_t
  1536. {
  1537. /** Number of bits for representing a column number */
  1538. static constexpr uint16_t IND_BITS = 10;
  1539. /** Set if the column of the field has been instantly dropped */
  1540. static constexpr uint16_t DROPPED = 1U << (IND_BITS + 5);
  1541. /** Set if the column was dropped and originally declared NOT NULL */
  1542. static constexpr uint16_t NOT_NULL = 1U << (IND_BITS + 4);
  1543. /** Column index (if !(data & DROPPED)): table->cols[data & IND],
  1544. or field length (if (data & DROPPED)):
  1545. (data & IND) = 0 if variable-length with max_len < 256 bytes;
  1546. (data & IND) = 1 if variable-length with max_len > 255 bytes;
  1547. (data & IND) = 1 + L otherwise, with L=fixed length of the column */
  1548. static constexpr uint16_t IND = (1U << IND_BITS) - 1;
  1549. /** Field metadata */
  1550. uint16_t data;
  1551. void clear_not_null() { data &= uint16_t(~NOT_NULL); }
  1552. public:
  1553. bool is_dropped() const { return data & DROPPED; }
  1554. void set_dropped() { data |= DROPPED; }
  1555. bool is_not_null() const { return data & NOT_NULL; }
  1556. void set_not_null() { ut_ad(is_dropped()); data |= NOT_NULL; }
  1557. uint16_t ind() const { return data & IND; }
  1558. void set_ind(uint16_t i)
  1559. {
  1560. DBUG_ASSERT(i <= IND);
  1561. DBUG_ASSERT(!ind());
  1562. data |= i;
  1563. }
  1564. field_map_element_t& operator= (uint16_t value)
  1565. {
  1566. data = value;
  1567. return *this;
  1568. }
  1569. operator uint16_t() { return data; }
  1570. };
  1571. static_assert(sizeof(field_map_element_t) == 2,
  1572. "Size mismatch for a persistent data item!");
  1573. /** Instantly dropped or reordered columns */
  1574. struct dict_instant_t
  1575. {
  1576. /** Number of dropped columns */
  1577. unsigned n_dropped;
  1578. /** Dropped columns */
  1579. dict_col_t* dropped;
  1580. /** Map of clustered index non-PK fields[i - first_user_field()]
  1581. to table columns */
  1582. field_map_element_t* field_map;
  1583. };
  1584. /** These are used when MySQL FRM and InnoDB data dictionary are
  1585. in inconsistent state. */
  1586. typedef enum {
  1587. DICT_FRM_CONSISTENT = 0, /*!< Consistent state */
  1588. DICT_FRM_NO_PK = 1, /*!< MySQL has no primary key
  1589. but InnoDB dictionary has
  1590. non-generated one. */
  1591. DICT_NO_PK_FRM_HAS = 2, /*!< MySQL has primary key but
  1592. InnoDB dictionary has not. */
  1593. DICT_FRM_INCONSISTENT_KEYS = 3 /*!< Key count mismatch */
  1594. } dict_frm_t;
  1595. /** Data structure for a database table. Most fields will be
  1596. initialized to 0, NULL or FALSE in dict_mem_table_create(). */
  1597. struct dict_table_t {
  1598. /** Get reference count.
  1599. @return current value of n_ref_count */
  1600. inline uint32_t get_ref_count() const { return n_ref_count; }
  1601. /** Acquire the table handle. */
  1602. inline void acquire();
  1603. /** Release the table handle.
  1604. @return whether the last handle was released */
  1605. inline bool release();
  1606. /** @return whether the table supports transactions */
  1607. bool no_rollback() const
  1608. {
  1609. return !(~unsigned(flags) & DICT_TF_MASK_NO_ROLLBACK);
  1610. }
  1611. /** @return whether this is a temporary table */
  1612. bool is_temporary() const
  1613. {
  1614. return flags2 & DICT_TF2_TEMPORARY;
  1615. }
  1616. /** @return whether the table is not in ROW_FORMAT=REDUNDANT */
  1617. bool not_redundant() const { return flags & DICT_TF_COMPACT; }
  1618. /** @return whether this table is readable
  1619. @retval true normally
  1620. @retval false if this is a single-table tablespace
  1621. and the .ibd file is missing, or a
  1622. page cannot be read or decrypted */
  1623. bool is_readable() const
  1624. {
  1625. ut_ad(file_unreadable || space);
  1626. return(UNIV_LIKELY(!file_unreadable));
  1627. }
  1628. /** @return whether the table is accessible */
  1629. bool is_accessible() const
  1630. {
  1631. return UNIV_LIKELY(is_readable() && !corrupted && space)
  1632. && !space->is_stopping();
  1633. }
  1634. /** Check if a table name contains the string "/#sql"
  1635. which denotes temporary or intermediate tables in MariaDB. */
  1636. static bool is_temporary_name(const char* name)
  1637. {
  1638. return strstr(name, "/" TEMP_FILE_PREFIX) != NULL;
  1639. }
  1640. /** @return whether instant ALTER TABLE is in effect */
  1641. bool is_instant() const
  1642. {
  1643. return(UT_LIST_GET_FIRST(indexes)->is_instant());
  1644. }
  1645. /** @return whether the table supports instant ALTER TABLE */
  1646. bool supports_instant() const
  1647. {
  1648. return(!(flags & DICT_TF_MASK_ZIP_SSIZE));
  1649. }
  1650. /** @return the number of instantly dropped columns */
  1651. unsigned n_dropped() const { return instant ? instant->n_dropped : 0; }
  1652. /** Look up an old column.
  1653. @param[in] cols the old columns of the table
  1654. @param[in] col_map map from old table columns to altered ones
  1655. @param[in] n_cols number of old columns
  1656. @param[in] i the number of the new column
  1657. @return old column
  1658. @retval NULL if column i was added to the table */
  1659. static const dict_col_t* find(const dict_col_t* cols,
  1660. const ulint* col_map, ulint n_cols,
  1661. ulint i)
  1662. {
  1663. for (ulint o = n_cols; o--; ) {
  1664. if (col_map[o] == i) {
  1665. return &cols[o];
  1666. }
  1667. }
  1668. return NULL;
  1669. }
  1670. /** Serialise metadata of dropped or reordered columns.
  1671. @param[in,out] heap memory heap for allocation
  1672. @param[out] field data field with the metadata */
  1673. inline void serialise_columns(mem_heap_t* heap, dfield_t* field) const;
  1674. /** Reconstruct dropped or reordered columns.
  1675. @param[in] metadata data from serialise_columns()
  1676. @param[in] len length of the metadata, in bytes
  1677. @return whether parsing the metadata failed */
  1678. bool deserialise_columns(const byte* metadata, ulint len);
  1679. /** Set is_instant() before instant_column().
  1680. @param[in] old previous table definition
  1681. @param[in] col_map map from old.cols[]
  1682. and old.v_cols[] to this
  1683. @param[out] first_alter_pos 0, or
  1684. 1 + first changed column position */
  1685. inline void prepare_instant(const dict_table_t& old,
  1686. const ulint* col_map,
  1687. unsigned& first_alter_pos);
  1688. /** Adjust table metadata for instant ADD/DROP/reorder COLUMN.
  1689. @param[in] table table on which prepare_instant() was invoked
  1690. @param[in] col_map mapping from cols[] and v_cols[] to table
  1691. @return whether the metadata record must be updated */
  1692. inline bool instant_column(const dict_table_t& table,
  1693. const ulint* col_map);
  1694. /** Roll back instant_column().
  1695. @param[in] old_n_cols original n_cols
  1696. @param[in] old_cols original cols
  1697. @param[in] old_col_names original col_names
  1698. @param[in] old_instant original instant structure
  1699. @param[in] old_fields original fields
  1700. @param[in] old_n_fields original number of fields
  1701. @param[in] old_n_core_fields original number of core fields
  1702. @param[in] old_n_v_cols original n_v_cols
  1703. @param[in] old_v_cols original v_cols
  1704. @param[in] old_v_col_names original v_col_names
  1705. @param[in] col_map column map */
  1706. inline void rollback_instant(
  1707. unsigned old_n_cols,
  1708. dict_col_t* old_cols,
  1709. const char* old_col_names,
  1710. dict_instant_t* old_instant,
  1711. dict_field_t* old_fields,
  1712. unsigned old_n_fields,
  1713. unsigned old_n_core_fields,
  1714. unsigned old_n_v_cols,
  1715. dict_v_col_t* old_v_cols,
  1716. const char* old_v_col_names,
  1717. const ulint* col_map);
  1718. /** Add the table definition to the data dictionary cache */
  1719. void add_to_cache();
  1720. /** @return whether the table is versioned.
  1721. It is assumed that both vers_start and vers_end set to 0
  1722. iff table is not versioned. In any other case,
  1723. these fields correspond to actual positions in cols[]. */
  1724. bool versioned() const { return vers_start || vers_end; }
  1725. bool versioned_by_id() const
  1726. {
  1727. return versioned() && cols[vers_start].mtype == DATA_INT;
  1728. }
  1729. void inc_fk_checks()
  1730. {
  1731. #ifdef UNIV_DEBUG
  1732. int32_t fk_checks=
  1733. #endif
  1734. n_foreign_key_checks_running++;
  1735. ut_ad(fk_checks >= 0);
  1736. }
  1737. void dec_fk_checks()
  1738. {
  1739. #ifdef UNIV_DEBUG
  1740. int32_t fk_checks=
  1741. #endif
  1742. n_foreign_key_checks_running--;
  1743. ut_ad(fk_checks > 0);
  1744. }
  1745. /** For overflow fields returns potential max length stored inline */
  1746. inline size_t get_overflow_field_local_len() const;
  1747. /** Parse the table file name into table name and database name.
  1748. @tparam dict_locked whether dict_sys.mutex is being held
  1749. @param[in,out] db_name database name buffer
  1750. @param[in,out] tbl_name table name buffer
  1751. @param[out] db_name_len database name length
  1752. @param[out] tbl_name_len table name length
  1753. @return whether the table name is visible to SQL */
  1754. template<bool dict_locked= false>
  1755. bool parse_name(char (&db_name)[NAME_LEN + 1],
  1756. char (&tbl_name)[NAME_LEN + 1],
  1757. size_t *db_name_len, size_t *tbl_name_len) const;
  1758. private:
  1759. /** Initialize instant->field_map.
  1760. @param[in] table table definition to copy from */
  1761. inline void init_instant(const dict_table_t& table);
  1762. public:
  1763. /** Id of the table. */
  1764. table_id_t id;
  1765. /** Hash chain node. */
  1766. hash_node_t id_hash;
  1767. /** Table name. */
  1768. table_name_t name;
  1769. /** Hash chain node. */
  1770. hash_node_t name_hash;
  1771. /** Memory heap */
  1772. mem_heap_t* heap;
  1773. /** NULL or the directory path specified by DATA DIRECTORY. */
  1774. char* data_dir_path;
  1775. /** The tablespace of the table */
  1776. fil_space_t* space;
  1777. /** Tablespace ID */
  1778. ulint space_id;
  1779. /** Stores information about:
  1780. 1 row format (redundant or compact),
  1781. 2 compressed page size (zip shift size),
  1782. 3 whether using atomic blobs,
  1783. 4 whether the table has been created with the option DATA DIRECTORY.
  1784. Use DICT_TF_GET_COMPACT(), DICT_TF_GET_ZIP_SSIZE(),
  1785. DICT_TF_HAS_ATOMIC_BLOBS() and DICT_TF_HAS_DATA_DIR() to parse this
  1786. flag. */
  1787. unsigned flags:DICT_TF_BITS;
  1788. /** Stores information about:
  1789. 1 whether the table has been created using CREATE TEMPORARY TABLE,
  1790. 2 whether the table has an internally defined DOC ID column,
  1791. 3 whether the table has a FTS index,
  1792. 4 whether DOC ID column need to be added to the FTS index,
  1793. 5 whether the table is being created its own tablespace,
  1794. 6 whether the table has been DISCARDed,
  1795. 7 whether the aux FTS tables names are in hex.
  1796. Use DICT_TF2_FLAG_IS_SET() to parse this flag. */
  1797. unsigned flags2:DICT_TF2_BITS;
  1798. /** TRUE if the table is an intermediate table during copy alter
  1799. operation or a partition/subpartition which is required for copying
  1800. data and skip the undo log for insertion of row in the table.
  1801. This variable will be set and unset during extra(), or during the
  1802. process of altering partitions */
  1803. unsigned skip_alter_undo:1;
  1804. /*!< whether this is in a single-table tablespace and the .ibd
  1805. file is missing or page decryption failed and page is corrupted */
  1806. unsigned file_unreadable:1;
  1807. /** TRUE if the table object has been added to the dictionary cache. */
  1808. unsigned cached:1;
  1809. /** TRUE if the table is to be dropped, but not yet actually dropped
  1810. (could in the background drop list). It is turned on at the beginning
  1811. of row_drop_table_for_mysql() and turned off just before we start to
  1812. update system tables for the drop. It is protected by dict_sys.latch. */
  1813. unsigned to_be_dropped:1;
  1814. /** Number of non-virtual columns defined so far. */
  1815. unsigned n_def:10;
  1816. /** Number of non-virtual columns. */
  1817. unsigned n_cols:10;
  1818. /** Number of total columns (inlcude virtual and non-virtual) */
  1819. unsigned n_t_cols:10;
  1820. /** Number of total columns defined so far. */
  1821. unsigned n_t_def:10;
  1822. /** Number of virtual columns defined so far. */
  1823. unsigned n_v_def:10;
  1824. /** Number of virtual columns. */
  1825. unsigned n_v_cols:10;
  1826. /** 1 + the position of autoinc counter field in clustered
  1827. index, or 0 if there is no persistent AUTO_INCREMENT column in
  1828. the table. */
  1829. unsigned persistent_autoinc:10;
  1830. /** TRUE if it's not an InnoDB system table or a table that has no FK
  1831. relationships. */
  1832. unsigned can_be_evicted:1;
  1833. /** TRUE if table is corrupted. */
  1834. unsigned corrupted:1;
  1835. /** TRUE if some indexes should be dropped after ONLINE_INDEX_ABORTED
  1836. or ONLINE_INDEX_ABORTED_DROPPED. */
  1837. unsigned drop_aborted:1;
  1838. /** Array of column descriptions. */
  1839. dict_col_t* cols;
  1840. /** Array of virtual column descriptions. */
  1841. dict_v_col_t* v_cols;
  1842. /** List of stored column descriptions. It is used only for foreign key
  1843. check during create table and copy alter operations.
  1844. During copy alter, s_cols list is filled during create table operation
  1845. and need to preserve till rename table operation. That is the
  1846. reason s_cols is a part of dict_table_t */
  1847. dict_s_col_list* s_cols;
  1848. /** Instantly dropped or reordered columns, or NULL if none */
  1849. dict_instant_t* instant;
  1850. /** Column names packed in a character string
  1851. "name1\0name2\0...nameN\0". Until the string contains n_cols, it will
  1852. be allocated from a temporary heap. The final string will be allocated
  1853. from table->heap. */
  1854. const char* col_names;
  1855. /** Virtual column names */
  1856. const char* v_col_names;
  1857. unsigned vers_start:10;
  1858. /*!< System Versioning: row start col index */
  1859. unsigned vers_end:10;
  1860. /*!< System Versioning: row end col index */
  1861. bool is_system_db;
  1862. /*!< True if the table belongs to a system
  1863. database (mysql, information_schema or
  1864. performance_schema) */
  1865. dict_frm_t dict_frm_mismatch;
  1866. /*!< !DICT_FRM_CONSISTENT==0 if data
  1867. dictionary information and
  1868. MySQL FRM information mismatch. */
  1869. /** The FTS_DOC_ID_INDEX, or NULL if no fulltext indexes exist */
  1870. dict_index_t* fts_doc_id_index;
  1871. /** List of indexes of the table. */
  1872. UT_LIST_BASE_NODE_T(dict_index_t) indexes;
  1873. #ifdef BTR_CUR_HASH_ADAPT
  1874. /** List of detached indexes that are waiting to be freed along with
  1875. the last adaptive hash index entry.
  1876. Protected by autoinc_mutex (sic!) */
  1877. UT_LIST_BASE_NODE_T(dict_index_t) freed_indexes;
  1878. #endif /* BTR_CUR_HASH_ADAPT */
  1879. /** List of foreign key constraints in the table. These refer to
  1880. columns in other tables. */
  1881. UT_LIST_BASE_NODE_T(dict_foreign_t) foreign_list;
  1882. /** List of foreign key constraints which refer to this table. */
  1883. UT_LIST_BASE_NODE_T(dict_foreign_t) referenced_list;
  1884. /** Node of the LRU list of tables. */
  1885. UT_LIST_NODE_T(dict_table_t) table_LRU;
  1886. /** Maximum recursive level we support when loading tables chained
  1887. together with FK constraints. If exceeds this level, we will stop
  1888. loading child table into memory along with its parent table. */
  1889. unsigned fk_max_recusive_level:8;
  1890. /** Count of how many foreign key check operations are currently being
  1891. performed on the table. We cannot drop the table while there are
  1892. foreign key checks running on it. */
  1893. Atomic_counter<int32_t> n_foreign_key_checks_running;
  1894. /** Transactions whose view low limit is greater than this number are
  1895. not allowed to store to the MySQL query cache or retrieve from it.
  1896. When a trx with undo logs commits, it sets this to the value of the
  1897. transaction id. */
  1898. trx_id_t query_cache_inv_trx_id;
  1899. /** Transaction id that last touched the table definition. Either when
  1900. loading the definition or CREATE TABLE, or ALTER TABLE (prepare,
  1901. commit, and rollback phases). */
  1902. trx_id_t def_trx_id;
  1903. /*!< set of foreign key constraints in the table; these refer to
  1904. columns in other tables */
  1905. dict_foreign_set foreign_set;
  1906. /*!< set of foreign key constraints which refer to this table */
  1907. dict_foreign_set referenced_set;
  1908. /** Statistics for query optimization. Mostly protected by
  1909. dict_sys.mutex. @{ */
  1910. /** TRUE if statistics have been calculated the first time after
  1911. database startup or table creation. */
  1912. unsigned stat_initialized:1;
  1913. /** Timestamp of last recalc of the stats. */
  1914. time_t stats_last_recalc;
  1915. /** The two bits below are set in the 'stat_persistent' member. They
  1916. have the following meaning:
  1917. 1. _ON=0, _OFF=0, no explicit persistent stats setting for this table,
  1918. the value of the global srv_stats_persistent is used to determine
  1919. whether the table has persistent stats enabled or not
  1920. 2. _ON=0, _OFF=1, persistent stats are explicitly disabled for this
  1921. table, regardless of the value of the global srv_stats_persistent
  1922. 3. _ON=1, _OFF=0, persistent stats are explicitly enabled for this
  1923. table, regardless of the value of the global srv_stats_persistent
  1924. 4. _ON=1, _OFF=1, not allowed, we assert if this ever happens. */
  1925. #define DICT_STATS_PERSISTENT_ON (1 << 1)
  1926. #define DICT_STATS_PERSISTENT_OFF (1 << 2)
  1927. /** Indicates whether the table uses persistent stats or not. See
  1928. DICT_STATS_PERSISTENT_ON and DICT_STATS_PERSISTENT_OFF. */
  1929. ib_uint32_t stat_persistent;
  1930. /** The two bits below are set in the 'stats_auto_recalc' member. They
  1931. have the following meaning:
  1932. 1. _ON=0, _OFF=0, no explicit auto recalc setting for this table, the
  1933. value of the global srv_stats_persistent_auto_recalc is used to
  1934. determine whether the table has auto recalc enabled or not
  1935. 2. _ON=0, _OFF=1, auto recalc is explicitly disabled for this table,
  1936. regardless of the value of the global srv_stats_persistent_auto_recalc
  1937. 3. _ON=1, _OFF=0, auto recalc is explicitly enabled for this table,
  1938. regardless of the value of the global srv_stats_persistent_auto_recalc
  1939. 4. _ON=1, _OFF=1, not allowed, we assert if this ever happens. */
  1940. #define DICT_STATS_AUTO_RECALC_ON (1 << 1)
  1941. #define DICT_STATS_AUTO_RECALC_OFF (1 << 2)
  1942. /** Indicates whether the table uses automatic recalc for persistent
  1943. stats or not. See DICT_STATS_AUTO_RECALC_ON and
  1944. DICT_STATS_AUTO_RECALC_OFF. */
  1945. ib_uint32_t stats_auto_recalc;
  1946. /** The number of pages to sample for this table during persistent
  1947. stats estimation. If this is 0, then the value of the global
  1948. srv_stats_persistent_sample_pages will be used instead. */
  1949. ulint stats_sample_pages;
  1950. /** Approximate number of rows in the table. We periodically calculate
  1951. new estimates. */
  1952. ib_uint64_t stat_n_rows;
  1953. /** Approximate clustered index size in database pages. */
  1954. ulint stat_clustered_index_size;
  1955. /** Approximate size of other indexes in database pages. */
  1956. ulint stat_sum_of_other_index_sizes;
  1957. /** How many rows are modified since last stats recalc. When a row is
  1958. inserted, updated, or deleted, we add 1 to this number; we calculate
  1959. new estimates for the table and the indexes if the table has changed
  1960. too much, see dict_stats_update_if_needed(). The counter is reset
  1961. to zero at statistics calculation. This counter is not protected by
  1962. any latch, because this is only used for heuristics. */
  1963. ib_uint64_t stat_modified_counter;
  1964. /** Background stats thread is not working on this table. */
  1965. #define BG_STAT_NONE 0
  1966. /** Set in 'stats_bg_flag' when the background stats code is working
  1967. on this table. The DROP TABLE code waits for this to be cleared before
  1968. proceeding. */
  1969. #define BG_STAT_IN_PROGRESS (1 << 0)
  1970. /** Set in 'stats_bg_flag' when DROP TABLE starts waiting on
  1971. BG_STAT_IN_PROGRESS to be cleared. The background stats thread will
  1972. detect this and will eventually quit sooner. */
  1973. #define BG_STAT_SHOULD_QUIT (1 << 1)
  1974. /** The state of the background stats thread wrt this table.
  1975. See BG_STAT_NONE, BG_STAT_IN_PROGRESS and BG_STAT_SHOULD_QUIT.
  1976. Writes are covered by dict_sys.mutex. Dirty reads are possible. */
  1977. byte stats_bg_flag;
  1978. bool stats_error_printed;
  1979. /*!< Has persistent stats error beein
  1980. already printed for this table ? */
  1981. /* @} */
  1982. /** AUTOINC related members. @{ */
  1983. /* The actual collection of tables locked during AUTOINC read/write is
  1984. kept in trx_t. In order to quickly determine whether a transaction has
  1985. locked the AUTOINC lock we keep a pointer to the transaction here in
  1986. the 'autoinc_trx' member. This is to avoid acquiring the
  1987. lock_sys_t::mutex and scanning the vector in trx_t.
  1988. When an AUTOINC lock has to wait, the corresponding lock instance is
  1989. created on the trx lock heap rather than use the pre-allocated instance
  1990. in autoinc_lock below. */
  1991. /** A buffer for an AUTOINC lock for this table. We allocate the
  1992. memory here so that individual transactions can get it and release it
  1993. without a need to allocate space from the lock heap of the trx:
  1994. otherwise the lock heap would grow rapidly if we do a large insert
  1995. from a select. */
  1996. lock_t* autoinc_lock;
  1997. /** Mutex protecting the autoinc counter and freed_indexes. */
  1998. std::mutex autoinc_mutex;
  1999. /** Autoinc counter value to give to the next inserted row. */
  2000. ib_uint64_t autoinc;
  2001. /** This counter is used to track the number of granted and pending
  2002. autoinc locks on this table. This value is set after acquiring the
  2003. lock_sys_t::mutex but we peek the contents to determine whether other
  2004. transactions have acquired the AUTOINC lock or not. Of course only one
  2005. transaction can be granted the lock but there can be multiple
  2006. waiters. */
  2007. ulong n_waiting_or_granted_auto_inc_locks;
  2008. /** The transaction that currently holds the the AUTOINC lock on this
  2009. table. Protected by lock_sys.mutex. */
  2010. const trx_t* autoinc_trx;
  2011. /* @} */
  2012. /** FTS specific state variables. */
  2013. fts_t* fts;
  2014. /** Quiescing states, protected by the dict_index_t::lock. ie. we can
  2015. only change the state if we acquire all the latches (dict_index_t::lock)
  2016. in X mode of this table's indexes. */
  2017. ib_quiesce_t quiesce;
  2018. /** Count of the number of record locks on this table. We use this to
  2019. determine whether we can evict the table from the dictionary cache.
  2020. It is protected by lock_sys.mutex. */
  2021. ulint n_rec_locks;
  2022. private:
  2023. /** Count of how many handles are opened to this table. Dropping of the
  2024. table is NOT allowed until this count gets to zero. MySQL does NOT
  2025. itself check the number of open handles at DROP. */
  2026. Atomic_counter<uint32_t> n_ref_count;
  2027. public:
  2028. /** List of locks on the table. Protected by lock_sys.mutex. */
  2029. table_lock_list_t locks;
  2030. /** Timestamp of the last modification of this table. */
  2031. time_t update_time;
  2032. #ifdef UNIV_DEBUG
  2033. /** Value of 'magic_n'. */
  2034. #define DICT_TABLE_MAGIC_N 76333786
  2035. /** Magic number. */
  2036. ulint magic_n;
  2037. #endif /* UNIV_DEBUG */
  2038. /** mysql_row_templ_t for base columns used for compute the virtual
  2039. columns */
  2040. dict_vcol_templ_t* vc_templ;
  2041. /* @return whether the table has any other transcation lock
  2042. other than the given transaction */
  2043. bool has_lock_other_than(const trx_t *trx) const
  2044. {
  2045. for (lock_t *lock= UT_LIST_GET_FIRST(locks); lock;
  2046. lock= UT_LIST_GET_NEXT(un_member.tab_lock.locks, lock))
  2047. if (lock->trx != trx)
  2048. return true;
  2049. return false;
  2050. }
  2051. /** Check whether the table name is same as mysql/innodb_stats_table
  2052. or mysql/innodb_index_stats.
  2053. @return true if the table name is same as stats table */
  2054. bool is_stats_table() const;
  2055. /** @return number of unique columns in FTS_DOC_ID index */
  2056. unsigned fts_n_uniq() const { return versioned() ? 2 : 1; }
  2057. /** @return the index for that starts with a specific column */
  2058. dict_index_t *get_index(const dict_col_t &col) const;
  2059. };
  2060. inline void dict_index_t::set_modified(mtr_t& mtr) const
  2061. {
  2062. mtr.set_named_space(table->space);
  2063. }
  2064. inline bool table_name_t::is_temporary() const
  2065. {
  2066. return dict_table_t::is_temporary_name(m_name);
  2067. }
  2068. inline bool dict_index_t::is_readable() const { return table->is_readable(); }
  2069. inline bool dict_index_t::is_instant() const
  2070. {
  2071. ut_ad(n_core_fields > 0);
  2072. ut_ad(n_core_fields <= n_fields || table->n_dropped());
  2073. ut_ad(n_core_fields == n_fields
  2074. || (type & ~(DICT_UNIQUE | DICT_CORRUPT)) == DICT_CLUSTERED);
  2075. ut_ad(n_core_fields == n_fields || table->supports_instant());
  2076. ut_ad(n_core_fields == n_fields || !table->is_temporary());
  2077. ut_ad(!table->instant || !table->is_temporary());
  2078. return n_core_fields != n_fields
  2079. || (is_primary() && table->instant);
  2080. }
  2081. inline bool dict_index_t::is_corrupted() const
  2082. {
  2083. return UNIV_UNLIKELY(online_status >= ONLINE_INDEX_ABORTED
  2084. || (type & DICT_CORRUPT)
  2085. || (table && table->corrupted));
  2086. }
  2087. inline void dict_index_t::clear_instant_add()
  2088. {
  2089. DBUG_ASSERT(is_primary());
  2090. DBUG_ASSERT(is_instant());
  2091. DBUG_ASSERT(!table->instant);
  2092. for (unsigned i= n_core_fields; i < n_fields; i++)
  2093. fields[i].col->clear_instant();
  2094. n_core_fields= n_fields;
  2095. n_core_null_bytes= static_cast<byte>
  2096. (UT_BITS_IN_BYTES(static_cast<unsigned>(n_nullable)));
  2097. }
  2098. inline void dict_index_t::clear_instant_alter()
  2099. {
  2100. DBUG_ASSERT(is_primary());
  2101. DBUG_ASSERT(n_fields == n_def);
  2102. if (!table->instant) {
  2103. if (is_instant()) {
  2104. clear_instant_add();
  2105. }
  2106. return;
  2107. }
  2108. #ifndef DBUG_OFF
  2109. for (unsigned i = first_user_field(); i--; ) {
  2110. DBUG_ASSERT(!fields[i].col->is_dropped());
  2111. DBUG_ASSERT(!fields[i].col->is_nullable());
  2112. }
  2113. #endif
  2114. const dict_col_t* ai_col = table->persistent_autoinc
  2115. ? fields[table->persistent_autoinc - 1].col
  2116. : NULL;
  2117. dict_field_t* const begin = &fields[first_user_field()];
  2118. dict_field_t* end = &fields[n_fields];
  2119. for (dict_field_t* d = begin; d < end; ) {
  2120. /* Move fields for dropped columns to the end. */
  2121. if (!d->col->is_dropped()) {
  2122. d++;
  2123. } else {
  2124. if (d->col->is_nullable()) {
  2125. n_nullable--;
  2126. }
  2127. std::swap(*d, *--end);
  2128. }
  2129. }
  2130. DBUG_ASSERT(&fields[n_fields - table->n_dropped()] == end);
  2131. n_core_fields = n_fields = n_def
  2132. = static_cast<unsigned>(end - fields) & MAX_N_FIELDS;
  2133. n_core_null_bytes = static_cast<byte>(UT_BITS_IN_BYTES(n_nullable));
  2134. std::sort(begin, end, [](const dict_field_t& a, const dict_field_t& b)
  2135. { return a.col->ind < b.col->ind; });
  2136. table->instant = NULL;
  2137. if (ai_col) {
  2138. auto a = std::find_if(fields, end,
  2139. [ai_col](const dict_field_t& f)
  2140. { return f.col == ai_col; });
  2141. table->persistent_autoinc = (a == end)
  2142. ? 0
  2143. : (1 + static_cast<unsigned>(a - fields))
  2144. & MAX_N_FIELDS;
  2145. }
  2146. }
  2147. /** @return whether the column was instantly dropped
  2148. @param[in] index the clustered index */
  2149. inline bool dict_col_t::is_dropped(const dict_index_t& index) const
  2150. {
  2151. DBUG_ASSERT(index.is_primary());
  2152. DBUG_ASSERT(!is_dropped() == !index.table->instant);
  2153. DBUG_ASSERT(!is_dropped() || (this >= index.table->instant->dropped
  2154. && this < index.table->instant->dropped
  2155. + index.table->instant->n_dropped));
  2156. return is_dropped();
  2157. }
  2158. /*******************************************************************//**
  2159. Initialise the table lock list. */
  2160. void
  2161. lock_table_lock_list_init(
  2162. /*======================*/
  2163. table_lock_list_t* locks); /*!< List to initialise */
  2164. /** A function object to add the foreign key constraint to the referenced set
  2165. of the referenced table, if it exists in the dictionary cache. */
  2166. struct dict_foreign_add_to_referenced_table {
  2167. void operator()(dict_foreign_t* foreign) const
  2168. {
  2169. if (dict_table_t* table = foreign->referenced_table) {
  2170. std::pair<dict_foreign_set::iterator, bool> ret
  2171. = table->referenced_set.insert(foreign);
  2172. ut_a(ret.second);
  2173. }
  2174. }
  2175. };
  2176. /** Check whether the col is used in spatial index or regular index.
  2177. @param[in] col column to check
  2178. @return spatial status */
  2179. inline
  2180. spatial_status_t
  2181. dict_col_get_spatial_status(
  2182. const dict_col_t* col)
  2183. {
  2184. spatial_status_t spatial_status = SPATIAL_NONE;
  2185. /* Column is not a part of any index. */
  2186. if (!col->ord_part) {
  2187. return(spatial_status);
  2188. }
  2189. if (DATA_GEOMETRY_MTYPE(col->mtype)) {
  2190. if (col->max_prefix == 0) {
  2191. spatial_status = SPATIAL_ONLY;
  2192. } else {
  2193. /* Any regular index on a geometry column
  2194. should have a prefix. */
  2195. spatial_status = SPATIAL_MIXED;
  2196. }
  2197. }
  2198. return(spatial_status);
  2199. }
  2200. /** Clear defragmentation summary. */
  2201. inline void dict_stats_empty_defrag_summary(dict_index_t* index)
  2202. {
  2203. index->stat_defrag_n_pages_freed = 0;
  2204. }
  2205. /** Clear defragmentation related index stats. */
  2206. inline void dict_stats_empty_defrag_stats(dict_index_t* index)
  2207. {
  2208. index->stat_defrag_modified_counter = 0;
  2209. index->stat_defrag_n_page_split = 0;
  2210. }
  2211. #include "dict0mem.inl"
  2212. #endif /* dict0mem_h */